2016P01 / cinematography
Niels Joubert, Jane L. E, Dan B Goldman, Floraine Berthouzoz, Mike Roberts, James A. Landay, Pat Hanrahan
Computes canonical shots of two people and transitions between them while respecting distance and vehicle constraints. RTK GPS and IMU trackers localize both actors and the camera drone.
Read findings, applications and limits
METHOD
Builds canonical shots from two-subject geometry and optimizes constrained transitions between paths.
EVALUATION
Evaluates outdoor filmed trajectories with position and orientation trackers carried by the subjects.
FINDING
Averaging two camera paths can violate separation; constrained blending addresses this.
PRACTICAL CONNECTION
Use screen-space placement to plan dialogue or ceremonial shots before selecting flight positions.
LIMITATION
Mainly considers one or two relatively stationary subjects. It is not a tag-free urban autonomy solution.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 1 Introduction · 7 Transitioning Between Shots · 8 Tracking and Control Platform · 9 Results · 11 Conclusion
DOI: 10.48550/arxiv.1610.01691
OpenAlex citations for this record: 65 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 65 ↗
Open primary text ↗ ·
Open primary text ↗2018P02 / cinematography
Quentin Galvane, Christophe Lino, Marc Christie, Julien Fleureau, Fabien Servant, François-Louis Tariolle, Philippe Guillotel
Introduces Drone Toric Space, a camera representation that connects composition to physical drone constraints, together with interactive path tools and coordinated multi-drone viewpoints.
Read findings, applications and limits
METHOD
Drone Toric Space encodes framing and vehicle restrictions for screen manipulation and sketched trajectories.
EVALUATION
Twelve novices compare manual, framing and sketch controls; physical experiments use Vicon and prepared obstacle maps.
FINDING
Composition, separation and camera actuation limits can be represented within one planning framework.
PRACTICAL CONNECTION
A useful foundation for interfaces that specify subject size and viewpoint before world coordinates.
LIMITATION
The experimental hardware limited camera tilt and stabilization; arbitrary outdoor operation was not established.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 4 Drone Toric Space · 8 Results · 8.5 Limitations · 9 Conclusion · 8.1 User evaluation
DOI: 10.1145/3181975
OpenAlex citations for this record: 64 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 64 ↗ · Crossref / separate-database: 60 ↗
Open primary text ↗ ·
Open primary text ↗2016P03 / cinematography
Quentin Galvane, Julien Fleureau, François-Louis Tariolle, Philippe Guillotel
Translates high-level shot descriptions into camera configurations and drone trajectories, adapting virtual cinematography techniques to physical UAV control.
Read findings, applications and limits
METHOD
Translates shot descriptions into viewpoints and trajectories before passing them to flight control.
EVALUATION
Physical evaluation uses an indoor motion-capture setup rather than autonomous exploration of unknown outdoors.
FINDING
Spherical and toric representations turn framing instructions into candidate viewpoints.
PRACTICAL CONNECTION
Use it to structure shot briefs by framing, subject and screen position.
LIMITATION
The paper explicitly lacks efficient obstacle avoidance; larger unprotected aircraft require further work.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 4 Autonomous path planning · 7 Experimental Results · 8 Limitations and Future work · 9 Conclusions
DOI: 10.2312/wiced.20161097
OpenAlex citations for this record: 29 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 29 ↗
Open primary text ↗ ·
Open primary text ↗2016P04 / cinematography
Julien Fleureau, Quentin Galvane, François-Louis Tariolle, Philippe Guillotel
Proposes a generic rotary-wing control architecture that connects cinematographic path planning to high-level UAV APIs through coupled dynamic models and feedback control.
Read findings, applications and limits
METHOD
A generic architecture maps cinematic pose commands to a vehicle model and feedback control.
EVALUATION
Single-ARDrone filming demonstrations illustrate the architecture; they do not validate multi-drone coordination.
FINDING
Separating shot specification from low-level vehicle control is the architectural contribution.
PRACTICAL CONNECTION
Separate common camera intentions from platform-specific adapters in multi-vendor software.
LIMITATION
Experiments were primarily single-drone and constrained by the AR.Drone camera and flight hardware.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 4 Drone and Cinematography · 6 Limitation and Future Work · 7 Conclusion
DOI: 10.1145/2935620.2935622
OpenAlex citations for this record: 34 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 34 ↗ · Crossref / separate-database: 27 ↗
Open primary text ↗ ·
Open primary text ↗2020P05 / cinematography
Rogerio Bonatti, Yanfu Zhang, Sanjiban Choudhury, Wenshan Wang, Sebastian Scherer
Balances smoothness, occlusion, obstacle clearance and artistic shot objectives through real-time trajectory optimization under noisy actor predictions.
Read findings, applications and limits
METHOD
A long-horizon gradient optimizer combines smoothness, obstacles, occlusion and desired-shot costs.
EVALUATION
Simulation and outdoor filming evaluate trajectories under prior-map and actor-GPS assumptions.
FINDING
Explicit occlusion costs improve visibility while sometimes increasing deviation from the preferred shot.
PRACTICAL CONNECTION
During forest tracking, check the camera-to-subject sightline as well as the flight corridor.
LIMITATION
This version relies on actor GPS and prior mapping; visual localization and online mapping were future work.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Motion planning formulation · 5 Experiments · 6 Conclusion and discussion
DOI: 10.1007/978-3-030-33950-0_11
OpenAlex citations for this record: 65 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 65 ↗ · Crossref / separate-database: 33 ↗
Open primary text ↗ ·
Open primary text ↗2019P06 / cinematography
Rogerio Bonatti, Cherie Ho, Wenshan Wang, Sanjiban Choudhury, Sebastian Scherer
Integrates visual actor localization, incremental LiDAR distance mapping and real-time camera motion planning to film moving targets in previously unknown environments.
Read findings, applications and limits
METHOD
Combines visual position and heading estimation, LiDAR mapping and onboard trajectory planning.
EVALUATION
DJI M210/Jetson TX2 trials film people and bicycles and compare noisy estimation and partial mapping.
FINDING
Actor heading estimation contributes to viewpoint selection, beyond simply knowing position.
PRACTICAL CONNECTION
Evaluate tracking, occlusion handling, mapping and replanning separately when comparing follow systems.
LIMITATION
Assumes a principally static environment; abrupt target changes and tracking failures can lose the subject.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · II Problem Formulation · System architecture · V Evaluation · VI Conclusion · V-A Experimental setup · V-B Field test results · Concluding limitations on static environments and reacquisition
DOI: 10.1109/iros40897.2019.8968163
OpenAlex citations for this record: 92 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 92 ↗ · Crossref / separate-database: 76 ↗
Open primary text ↗ ·
Open primary text ↗2020P07 / cinematography
Rogerio Bonatti, Wenshan Wang, Cherie Ho, Aayush Ahuja, Mirko Gschwindt, Efe Camci, Erdal Kayacan, Sanjiban Choudhury, Sebastian Scherer
Combines actor estimation, incremental distance mapping, trajectory optimization and learned artistic viewpoint selection in an integrated outdoor filming system.
Read findings, applications and limits
METHOD
Separates world-coordinate motion planning from image-feedback gimbal control, with learned shot selection.
EVALUATION
Outdoor system and module evaluations document cascading estimation errors and imperfect simulation transfer.
FINDING
The lessons learned emphasize cascading errors and separating image-based camera control from world-space planning.
PRACTICAL CONNECTION
Use the system breakdown to diagnose interactions between tracking errors, localization and jerky footage.
LIMITATION
Multi-drone, multi-actor extensions and richer models of artistic reasoning remain open.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 6 Mapping · 7 Motion Planning · 9 Experiments · 10.1 Lessons Learned · 11 Conclusion · 10.1 Lessons Learned: cascading error, height drift and simulation transfer
DOI: 10.1002/rob.21931
OpenAlex citations for this record: 91 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 91 ↗ · Crossref / separate-database: 92 ↗
Open primary text ↗ ·
Open primary text ↗2019P08 / cinematography
Mirko Gschwindt, Efe Camci, Rogerio Bonatti, Wenshan Wang, Erdal Kayacan, Sebastian Scherer
Uses deep reinforcement learning to select filming viewpoints, comparing handcrafted and human-derived rewards before testing transfer from simulation to a real drone.
Read findings, applications and limits
METHOD
A DQN selects left/right/front/back every six seconds from a height map, current shot and repetition count.
EVALUATION
Ten participants rank clips from five unseen scenes; the study does not test whole-film directing.
FINDING
Human preferences can inform viewpoint policies; losing the actor remains strongly undesirable.
PRACTICAL CONNECTION
Separate preference scores from visibility and safety criteria when designing shot recommendations.
LIMITATION
The policy selects among shot modes; it does not model complete film direction, narrative or editing.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Learning framework · User study · VII Conclusion · III-B State/action/reward formulation · V-C User study results · VI Discussion
DOI: 10.1109/iros40897.2019.8967592
OpenAlex citations for this record: 58 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 58 ↗ · Crossref / separate-database: 52 ↗
Open primary text ↗ ·
Open primary text ↗2020P09 / cinematography
Luis-Evaristo Caraballo, Ángel Montes-Romero, José-Miguel Díaz-Báñez, Jesús Capitán, Arturo Torres-González, Aníbal Ollero
Allocates filming tasks across UAVs to maximize recorded coverage while accounting for battery capacity and return to base.
Read findings, applications and limits
METHOD
Optimizes filming duration on a discrete graph, allocating single-vehicle solutions sequentially to a team.
EVALUATION
Randomized simulations and two-/three-UAV filming demonstrations assess scheduling, including battery replacement.
FINDING
A graph-based single-UAV solution is extended greedily to small teams.
PRACTICAL CONNECTION
Combine the shot list with travel, recharge and vehicle-change schedules for event coverage.
LIMITATION
The multi-UAV greedy solution is not globally optimal and simplifies energy consumption.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Problem formulation · IV Greedy strategy · Field experiments · VII Conclusions
DOI: 10.1109/iros45743.2020.9341622
OpenAlex citations for this record: 16 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 16 ↗ · Crossref / separate-database: 11 ↗
Open primary text ↗ ·
Open primary text ↗2024P10 / cinematography
Pablo Pueyo, Juan Dendarieta, Eduardo Montijano, Ana C. Murillo, Mac Schwager
Extends autonomous camera control to focus, aperture and focal length as well as camera pose, using perception and nonlinear model predictive control.
Read findings, applications and limits
METHOD
Nonlinear MPC jointly regulates framing, distance and depth of field through vehicle and lens settings.
EVALUATION
Flight experiment E6 films a chair without orientation/aperture control; E7 measures Jetson computation using recorded data.
FINDING
Joint intrinsic and extrinsic control enables effects beyond position-and-orientation planning.
PRACTICAL CONNECTION
Distinguish changes in image scale from changes in perspective when planning camera moves.
LIMITATION
The real-flight test did not control camera orientation and aperture; full-variable hardware capability should not be assumed.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · V Control Module · VIII Experiments · VIII-B Real Experiments · IX Future work and limitations · X Conclusions · VIII-B3 Experiment 6: real flight constraints · VIII-B4 Experiment 7: recorded-data compute test
DOI: 10.1109/tro.2024.3353550
OpenAlex citations for this record: 17 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 17 ↗ · Crossref / separate-database: 19 ↗
Open primary text ↗ ·
Open primary text ↗2023P11 / cinematography
Pablo Pueyo, Eduardo Montijano, Ana C. Murillo, Mac Schwager
Extracts subject composition and focus characteristics from a reference video and converts them into CineMPC instructions for a new scene.
Read findings, applications and limits
METHOD
Extracts pose, segmentation and focus features, smooths them by optimization, then issues CineMPC instructions.
EVALUATION
Uses two source sequences—one movie and one synthetic—and three virtual target environments.
FINDING
Style is represented through transferable framing and depth-of-field features.
PRACTICAL CONNECTION
Annotate subject placement and focus changes alongside camera movement in reference analysis.
LIMITATION
Real source videos were analyzed, but target filming was evaluated in simulation; real mobile-robot experiments remained future work.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Feature extraction and optimization · VI Experiments · VII Conclusions · VI-A Source sequences and target environments · VI-B Implementation details
DOI: 10.1109/iros55552.2023.10342280
OpenAlex citations for this record: 3 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 3 ↗ · Crossref / separate-database: 3 ↗
Open primary text ↗ ·
Open primary text ↗2019P12 / cinematography
Chong Huang, Yuanjie Dang, Peng Chen, Xin Yang, Kwang-Ting (Tim) Cheng
Learns style features and action imitation so that a single demonstration can guide filming without training a separate model for each demonstrated style.
Read findings, applications and limits
METHOD
Transfers a demonstration using representations pretrained on basic filming styles.
EVALUATION
Splits 146 videos into 97 training/49 test videos; evaluates motion error, classification, ten-person ratings and drone examples.
FINDING
One-shot transfer still relies on a model trained on basic filming styles.
PRACTICAL CONNECTION
Evaluate transfer to new subjects separately from transfer to new camera styles.
LIMITATION
The study focuses on a single person and filters difficult lighting, tiny subjects and prolonged occlusions from its data.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 3 One-Shot Imitation Filming · 5.1 Dataset · 5 Experiments · 6 Conclusions · 5.2 Experimental Setup and evaluation metrics
DOI: 10.48550/arxiv.1912.10609
OpenAlex citations for this record: 4 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 4 ↗ · OpenAlex / related-version-or-duplicate-not-added: 11 ↗
Open primary text ↗ ·
Open primary text ↗2019P13 / cinematography
Chong Huang, Chuan-En Lin, Zhenyu Yang, Yan Kong, Peng Chen, Xin Yang, Kwang-Ting Cheng
Uses subject skeletons, scene appearance and optical-flow camera motion in a sequence model trained from professional human-motion video clips.
Read findings, applications and limits
METHOD
Feeds skeleton, background and optical-flow cues into a sequence-to-sequence model to predict camera motion.
EVALUATION
Reports data evaluations across six filming styles and DJI Matrice 100 filming demonstrations.
FINDING
Temporal context and previous camera movement matter beyond a single attractive frame.
PRACTICAL CONNECTION
Review the opening, middle and closing composition together with speed changes.
LIMITATION
Evaluation uses six grouped styles and a custom platform, not every genre or environment.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Method and features · 4 Experiments · 4.3 Drone Cinematography System · 5 Conclusions
DOI: 10.1109/cvpr.2019.00437
OpenAlex citations for this record: 35 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 35 ↗ · Crossref / separate-database: 31 ↗
Open primary text ↗ ·
Open primary text ↗2020P14 / cinematography
Amirsaman Ashtari, Stefan Stevšić, Tobias Nägeli, Jean-Charles Bazin, Otmar Hilliges
Models the motion of a walking camera operator and camera rig so a drone can reproduce subjective first-person shots and transitions to smooth movement.
Read findings, applications and limits
METHOD
Models walking-operator and camera-rig motion, then converts it to feasible aerial camera movement.
EVALUATION
Combines real filming, comparison with walking motion-capture data and an online perceptual study with 106 respondents.
FINDING
Intentional camera shake can be optimized within aircraft dynamics rather than removed indiscriminately.
PRACTICAL CONNECTION
Choose deliberate motion character according to the scene rather than maximizing stabilization everywhere.
LIMITATION
Actuation and SDK limits prevent exact imitation of all human movements.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Motion model · Evaluation · 10 Limitations and Future Work · Conclusion · Perceptual user study · 9 Evaluation of our walking camera model
DOI: 10.1145/3378673
OpenAlex citations for this record: 25 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 25 ↗ · Crossref / separate-database: 25 ↗
Open primary text ↗ ·
Open primary text ↗2017P15 / cinematography
Tobias Nägeli, Lukas Meier, Alexander Domahidi, Javier Alonso-Mora, Otmar Hilliges
Uses virtual rails and framing objectives in receding-horizon optimization to jointly plan motion and control for one or multiple filming drones.
Read findings, applications and limits
METHOD
Receding-horizon control combines virtual rails, framing and sequential multi-drone visibility/collision coordination.
EVALUATION
Indoor tests use Vicon and Bebop2; evaluation includes computation scaling and preliminary feedback from one camera operator.
FINDING
Feedback and dynamic constraints are incorporated directly into planning rather than only trajectory tracking.
PRACTICAL CONNECTION
Describe storyboards with subject-relative intentions and tolerances, not only fixed waypoints.
LIMITATION
Feasibility is model- and horizon-dependent; optical focus and depth-of-field control were future extensions.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 3 Planning formulation · 4 Multi-Drone Flight · 5 Evaluation · 6 Conclusions · 5.1 Implementation Details · 5.2 Quantitative and qualitative experiments · 5.3 Preliminary expert feedback
DOI: 10.1145/3072959.3073712
OpenAlex citations for this record: 168 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 168 ↗ · Crossref / separate-database: 149 ↗
Open primary text ↗ ·
Open primary text ↗2019P16 / survey
Ioannis Mademlis, Nikos Nikolaidis, Anastasios Tefas, Ioannis Pitas, Tilman Wagner, Alberto Messina
Surveys autonomous aerial cinematography and formalizes framing, single-UAV motion and multi-UAV motion into a shot taxonomy linked to technical requirements.
Read findings, applications and limits
METHOD
Organizes elementary and combined camera motions alongside aircraft constraints and optical compensation.
EVALUATION
A tutorial categorizing 26 camera-motion types, not a hardware benchmark validating all types.
FINDING
Shot types differ in energy demands and autonomy complexity, not just visual appearance.
PRACTICAL CONNECTION
Organize a shot library with separate framing, aircraft-motion and camera-motion dimensions.
LIMITATION
A taxonomy and tutorial are not evidence that all shot types can be safely automated on arbitrary hardware.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Taxonomy of camera motion types · Focal-length and multi-drone compensation · Discussion
DOI: 10.1145/3347713
OpenAlex citations for this record: 49 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 49 ↗ · Crossref / separate-database: 44 ↗
Open primary text ↗ ·
Open primary text ↗2011P17 / motion
Daniel Mellinger, Vijay Kumar
Uses differential flatness and polynomial optimization to generate smooth trajectories through position/yaw keyframes and prescribed safe corridors.
Read findings, applications and limits
METHOD
Uses differential flatness and polynomial trajectories minimizing snap, the fourth time derivative of position.
EVALUATION
Hummingbird/Vicon experiments include hoop interception and constrained passages under external motion capture.
FINDING
Smooth higher derivatives connect geometric waypoints to dynamically trackable motion.
PRACTICAL CONNECTION
A foundation for understanding smooth takeoff, easing and turn transitions.
LIMITATION
Experiments use indoor motion capture; obstacle perception and cinematic composition are separate problems.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · II Model · V Trajectory Generation · VI Experiments · VI-A Spatially Scaled Trajectories
DOI: 10.1109/icra.2011.5980409
OpenAlex citations for this record: 2421 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 2421 ↗ · Crossref / separate-database: 2100 ↗
Open primary text ↗ ·
Open primary text ↗2016P18 / motion
Charles Richter, Adam Bry, Nicholas Roy
Develops numerically stable polynomial trajectory optimization and segment-time allocation for flight through cluttered indoor spaces.
Read findings, applications and limits
METHOD
Combines low-dimensional route search, numerically stable polynomial optimization and segment-time allocation.
EVALUATION
Numerical stability benchmarks and MIT indoor flights use onboard LiDAR, IMU and computation.
FINDING
Separating route search from polynomial optimization improves efficiency at the expense of global optimality.
PRACTICAL CONNECTION
Treat segment timing as a design variable in long waypoint sequences.
LIMITATION
Demonstrated aggressive-flight speeds are not recommended filming speeds or general safety limits.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Polynomial optimization formulation · Time allocation · 6 Conclusion · 4.2 Performance of Polynomial Optimization · 4.3 Experimental Flight Tests
DOI: 10.1007/978-3-319-28872-7_37
OpenAlex citations for this record: 764 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 764 ↗ · Crossref / separate-database: 565 ↗
Open primary text ↗ ·
Open primary text ↗2021P19 / motion
Xin Zhou, Zhepei Wang, Hongkai Ye, Chao Xu, Fei Gao
Avoids building a full ESDF by extracting obstacle information when a candidate trajectory collides, then optimizes and retimes the trajectory.
Read findings, applications and limits
METHOD
Derives obstacle repulsion from collision-guiding paths, optimizes B-splines and refines timing.
EVALUATION
Benchmarks and real indoor, abrupt-goal and forest trials assess navigation, not image composition.
FINDING
Targeted obstacle information reduces redundant computation during local planning.
PRACTICAL CONNECTION
Measure mapping and replanning latency separately in autonomous filming systems.
LIMITATION
It is a navigation planner, not a complete framing or cinematic-occlusion optimizer.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Planning framework · VI Experiment Results · VII Conclusion and Future Work · VI-A Implementation Details · VI-E Real-world Experiments
DOI: 10.1109/lra.2020.3047728
OpenAlex citations for this record: 557 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 557 ↗ · Crossref / separate-database: 568 ↗
Open primary text ↗ ·
Open primary text ↗2019P20 / motion
Boyu Zhou, Fei Gao, Luqi Wang, Chuhao Liu, Shaojie Shen
Combines kinodynamic search, B-spline trajectory optimization and time adjustment for efficient autonomous quadrotor navigation.
Read findings, applications and limits
METHOD
Kinodynamic search initializes B-spline optimization, followed by time reallocation for motion limits.
EVALUATION
Computational and physical obstacle-environment tests assess flight; cinematic occlusion and framing require additional design.
FINDING
Coarser spatial resolution trades computational efficiency against feasible-path coverage.
PRACTICAL CONNECTION
Consider map resolution, vehicle size and latency together for forest navigation.
LIMITATION
Large-scale dynamic scenes and swarms were future work; aesthetic quality is not the direct objective.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · VI Implementation Details · VII Results · VIII Conclusion
DOI: 10.1109/lra.2019.2927938
OpenAlex citations for this record: 754 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 754 ↗ · Crossref / separate-database: 749 ↗
Open primary text ↗ ·
Open primary text ↗2018P21 / motion
Davide Falanga, Philipp Foehn, Peng Lu, Davide Scaramuzza
Jointly optimizes vehicle motion and perception, keeping a point of interest visible and limiting its image-plane velocity.
Read findings, applications and limits
METHOD
MPC optimizes feature-centroid placement and image velocity jointly with vehicle motion.
EVALUATION
Experiments use a 420 g platform, two-second horizon and 100 Hz control; these are platform-specific conditions.
FINDING
Accelerating the aircraft can conflict with sensing objectives through required attitude changes.
PRACTICAL CONNECTION
Review image motion and aircraft tilt when tracking becomes unstable.
LIMITATION
Perception objectives are soft costs, not an unconditional guarantee that the target remains visible.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Optimization formulation · V Experiments · VI Discussion · VII Conclusions · VI Discussion: centroid, horizon and computation
DOI: 10.1109/iros.2018.8593739
OpenAlex citations for this record: 11 · 2026-10-21 · Source ↗
The same DOI is split between publication (11) and repository-derived (248) OpenAlex records. The displayed 11 is the DOI lookup result; Crossref reports 231. Counts are not added.
OpenAlex / primary-record: 11 ↗ · Crossref / separate-database: 231 ↗ · OpenAlex / related-version-or-duplicate-not-added: 248 ↗
Open primary text ↗ ·
Open primary text ↗2018P22 / perception
Tong Qin, Peiliang Li, Shaojie Shen
Combines monocular features and inertial measurements with initialization, failure recovery, relocalization and pose-graph optimization for metric state estimation.
Read findings, applications and limits
METHOD
Tightly couples preintegrated IMU and image features with relocalization and pose-graph correction.
EVALUATION
Evaluates EuRoC, indoor/outdoor trajectories and drone control; loop closure is disabled in the trajectory-tracking flight test.
FINDING
Sensor fusion improves estimation, but certain motions and environments remain weakly observable.
PRACTICAL CONNECTION
A foundation for understanding visual-inertial positioning when GNSS is weak.
LIMITATION
Calibration and visual conditions matter; feature-poor water cannot be assumed to match benchmark performance.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Estimator architecture · VIII Global Pose Graph Optimization · Experiments · X Conclusion and Future Work · IX Experiments: trajectory-tracking flight configuration
DOI: 10.1109/tro.2018.2853729
OpenAlex citations for this record: 4646 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 4646 ↗ · Crossref / separate-database: 4371 ↗
Open primary text ↗ ·
Open primary text ↗2017P23 / perception
Raúl Mur-Artal, Juan D. Tardós
Provides visual SLAM for monocular, stereo and RGB-D cameras with relocalization, loop closure and map reuse.
Read findings, applications and limits
METHOD
Uses ORB features for tracking, mapping and loop closure across monocular, stereo and RGB-D input.
EVALUATION
Benchmarks include KITTI, EuRoC and TUM RGB-D, with tracking loss reported under severe EuRoC motion blur.
FINDING
Metric scale in stereo/RGB-D operation should not be confused with monocular-only scale.
PRACTICAL CONNECTION
Useful background for repeatable camera routes and localization in previously mapped spaces.
LIMITATION
The evaluation reports tracking loss under severe motion blur; robustness is conditional.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · System overview · IV Evaluation · V Conclusion · IV-B EuRoC Dataset
DOI: 10.1109/tro.2017.2705103
OpenAlex citations for this record: 6886 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 6886 ↗ · Crossref / separate-database: 6272 ↗
Open primary text ↗ ·
Open primary text ↗2017P24 / perception
Helen Oleynikova, Zachary Taylor, Marius Fehr, Roland Siegwart, Juan Nieto
Builds incremental Euclidean signed distance fields from truncated distance fields for onboard trajectory planning and interpretable surface mapping.
Read findings, applications and limits
METHOD
Incrementally converts a depth-fused TSDF into an ESDF for obstacle-distance queries.
EVALUATION
Evaluates distance-field error, computation and MAV planning; voxel size and unknown-space policy affect behavior.
FINDING
Unknown space and map approximation error require explicit treatment in planning.
PRACTICAL CONNECTION
Check whether map resolution and sensing can represent thin branches or wires.
LIMITATION
Finite voxels and sensor errors mean a mapped gap is not automatically a safe physical corridor.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Mapping representation · VII MAV Planning Experiments · VIII Conclusions
DOI: 10.1109/iros.2017.8202315
OpenAlex citations for this record: 727 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 727 ↗ · Crossref / separate-database: 681 ↗
Open primary text ↗ ·
Open primary text ↗2020P25 / perception
Zachary Teed, Jia Deng
Estimates dense optical flow through all-pairs feature correlations and recurrent refinement, with strong benchmark accuracy at publication.
Read findings, applications and limits
METHOD
Builds multiscale all-pairs correlations and iteratively updates two-frame optical flow.
EVALUATION
Sintel/KITTI benchmarks and architecture/iteration ablations evaluate flow rather than physical flight control.
FINDING
Image displacement is not itself a metric 3D camera trajectory.
PRACTICAL CONNECTION
Analyze foreground/background flow when studying shake and perceived camera speed.
LIMITATION
It is not a flight controller; occlusion, blur and domain changes remain relevant limitations.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · 1 Introduction · Architecture · Experiments · Conclusions
DOI: 10.1007/978-3-030-58536-5_24
OpenAlex citations for this record: 2803 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 2803 ↗ · Crossref / separate-database: 2456 ↗ · OpenAlex / related-version-or-duplicate-not-added: 112 ↗
Open primary text ↗ ·
Open primary text ↗2019P26 / perception
Bo Li, Wei Wu, Qiang Wang, Fangyi Zhang, Junliang Xing, Junjie Yan
Improves Siamese visual tracking through deeper backbones, multi-layer feature aggregation and depthwise cross-correlation.
Read findings, applications and limits
METHOD
Improves Siamese tracking with layered features, depthwise correlation and spatially shifted training samples.
EVALUATION
Evaluates VOT, LaSOT, TrackingNet and aerial UAV123, without guaranteeing failure-free tracking.
FINDING
Detecting an object category and maintaining a particular target identity are distinct tasks.
PRACTICAL CONNECTION
Test identity retention around similar-looking subjects and temporary occlusion.
LIMITATION
General tracking benchmarks do not establish aircraft safety or performance in every aerial scene.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Network design · Experiments · 5 Conclusions · 4.4 Comparison with the state-of-the-art
DOI: 10.1109/cvpr.2019.00441
OpenAlex citations for this record: 2588 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 2588 ↗ · Crossref / separate-database: 2206 ↗
Open primary text ↗ ·
Open primary text ↗2017P27 / simulation
Shital Shah, Debadeepta Dey, Chris Lovett, Ashish Kapoor
Introduces a modular simulator for visual and physical vehicle testing, synthetic data and hardware-in-the-loop experimentation.
Read findings, applications and limits
METHOD
Separates vehicle physics, sensor simulation and rendering, with interfaces to real flight controllers.
EVALUATION
The architecture paper demonstrates examples and records then-missing collision and camera-model capabilities.
FINDING
Shared APIs help experimentation but do not eliminate the simulation-to-reality gap.
PRACTICAL CONNECTION
Evaluate both the rendered composition and the modeled vehicle motion during previsualization.
LIMITATION
The paper notes missing collision, lens and GNSS-degradation effects; current software maintenance needs separate verification.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Simulator architecture · Experiments · 5 Conclusion and Future Work
DOI: 10.1007/978-3-319-67361-5_40
OpenAlex citations for this record: 2247 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 2247 ↗ · Crossref / separate-database: 1650 ↗ · OpenAlex / related-version-or-duplicate-not-added: 26 ↗
Open primary text ↗ ·
Open primary text ↗2021P28 / simulation
Yunlong Song, Selim Naji, Elia Kaufmann, Antonio Loquercio, Davide Scaramuzza
Separates rendering and dynamics so simulation speed and fidelity can be configured independently for learning, navigation and interaction.
Read findings, applications and limits
METHOD
Decouples Unity rendering from dynamics for parallel multi-vehicle learning workloads.
EVALUATION
Measures dynamics and rendering separately; rendering rate depends on image size, scene and vehicle count.
FINDING
Visual realism and dynamic fidelity are separate dimensions of simulator quality.
PRACTICAL CONNECTION
Choose fidelity settings separately for visual shot previews and controller training.
LIMITATION
Reported physics throughput is not simultaneous photorealistic rendering speed or proof of real-flight transfer.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Simulation architecture · Experiments · Conclusion and Discussion · 4.1 Simulation Speed
DOI: 10.48550/arxiv.2009.00563
OpenAlex citations for this record: 137 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 137 ↗
Open primary text ↗ ·
Open primary text ↗2010P29 / motion
Taeyoung Lee, Melvin Leok, N. Harris McClamroch
Develops geometric tracking controllers on SE(3) for attitude, position and velocity flight modes without Euler-angle singularities.
Read findings, applications and limits
METHOD
Designs attitude, position and velocity control directly on SE(3), with stability analysis.
EVALUATION
Validation in this manuscript is numerical simulation, not physical reliability testing outside the proof assumptions.
FINDING
Four thrust inputs control a six-degree-of-freedom vehicle, coupling translation and attitude.
PRACTICAL CONNECTION
Distinguish aircraft attitude control from the gimbal’s role in maintaining the picture.
LIMITATION
Stability results and numerical demonstrations rely on model assumptions, not arbitrary disturbances or failures.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Dynamic model · Tracking controllers · Numerical examples · VIII Conclusions
DOI: 10.48550/arxiv.1003.2005
OpenAlex citations for this record: 104 · 2026-10-21 · Source ↗
The 104 citations belong to this preprint title; counts from the differently titled Geometric Tracking Control paper are not substituted.
OpenAlex / primary-record: 104 ↗
Open primary text ↗ ·
Open primary text ↗2021P30 / cinematography
Alfonso Alcántara, Jesús Capitán, Rita Cunha, Aníbal Ollero
Plans coordinated multi-UAV filming trajectories with vehicle dynamics, collision constraints, gimbal limits, smoothness and mutual camera visibility.
Read findings, applications and limits
METHOD
Priority-based distributed optimization handles motion, gimbal bounds, smoothness and mutual camera visibility.
EVALUATION
Uses simulation and three-UAV outdoor filming of one actor; arbitrary fleet sizes and networks remain unvalidated.
FINDING
Inter-drone separation and keeping other aircraft out of the image require different constraints.
PRACTICAL CONNECTION
Check airspace separation and image coverage together in multi-camera plans.
LIMITATION
Priority-based distributed planning does not guarantee a globally optimal team solution.
READING SCOPE
Full text retrieved; the following key sections were reviewed. Equations, code and experiments were not independently reproduced.
Abstract · Optimization formulation · Simulation and field experiments · 7 Conclusions · 6 Field Experiments
DOI: 10.1016/j.robot.2021.103778
OpenAlex citations for this record: 56 · 2026-10-21 · Source ↗
Count for this OpenAlex record; versions and duplicate records are not summed, and no global rank is claimed.
OpenAlex / primary-record: 56 ↗ · Crossref / separate-database: 54 ↗
Open primary text ↗ ·
Open primary text ↗