RADAR Perception for Dynamic Obstacle Avoidance onboard small-scale Quadrotor UAVs

Bernhard Rinner; Dnyandeep Mandaokar

REVIEW 4 major objections 5 minor 22 references

RADAR Perception for Dynamic Obstacle Avoidance onboard small-scale Quadrotor UAVs

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Millimeter-wave radar plus a barrier-function controller lets a small quadrotor dodge fast balls in darkness and smoke.

desk verdict Real radar perception data and a measured 14 ms loop; just don't trust the 89% avoidance rate until the HiL plant model is disclosed. read the letter →

arxiv 2608.01855 v1 pith:UVEDM34G submitted 2026-08-03 cs.RO cs.SYeess.SY

Dnyandeep Mandaokar , Bernhard Rinner This is my paper

classification cs.ROcs.SYeess.SY

keywords millimeter-waveradardynamicobstacleavoidancequadrotorcontrolbarrierfunctioninteractingmultiplemodeltrackerRaspberryPi4Bhardware-in-the-loopsmokeandlow-lightperception

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a small quadrotor can perform fast dynamic obstacle avoidance using only a millimeter-wave radar and a low-power onboard computer, without cameras, lidar, or GPUs. It derives a simple sufficient condition linking sensing range, relative speed, and total perception-to-command latency, and reports a 100 Hz perception, 200 Hz control system implemented on a Raspberry Pi 4B. In experiments with thrown balls of three sizes, the system achieved position errors within 0.15 m, 0.93 m, and 0.87 m in x, y, z, and an overall 89% avoidance rate at average ball speeds above 3.1 m/s, with an end-to-end latency of about 14 ms. If correct, this makes radar a viable primary sensing modality for fast obstacle avoidance in degraded visual conditions, opening a path to low-cost autonomous drones that work in smoke and darkness.

What carries the argument

The central object is the critical-distance inequality d_crit = v_rel (t_end + t_evade), which converts a timing budget (perception-to-command latency plus evasive-maneuver time) into a spatial trigger boundary. The system also relies on a 77–81 GHz FMCW radar configured for 100 Hz pointclouds, a sliding-window DBSCAN detector with radial-velocity filtering, a SAFE-IMM tracker using constant-velocity and constant-acceleration Kalman filters, and a DR-ACBF controller that maps noisy radar states to evasive accelerations via a Gauss-Southwell projection. The DR-CVaR trigger expands safety margins when the drone approaches the boundary, so the whole chain compensates for radar noise and actuati

What would settle it

A direct physical test: repeat the 390 throws with the actual quadrotor airborne (not tethered) and compare closest-approach distances and avoidance rate against the hardware-in-the-loop numbers; if real ds falls below the 0.8 m safety radius in a significant fraction of cases, the simulated-plant assumption fails. Also, a CPU spike that raises t_end to 25 ms on the Pi 4B would expand d_crit and should produce collisions, which the paper explicitly predicts.

Watch

Extended reading notes

Core claim

The central discovery is that a latency-bounded radar perception and control loop can meet the spatial requirements of fast dynamic obstacle avoidance on a resource-limited quadrotor. The paper establishes a critical distance d_crit = v_rel*(t_end + t_evade), the minimum range at which avoidance must be triggered so that the platform can laterally displace itself by the combined safety radius before the obstacle arrives. It then validates that a 77–81 GHz FMCW radar, a lightweight interacting-multiple-model tracker, and a distributionally robust acceleration control barrier filter that outputs evasive accelerations directly achieve real-time operation: 100 Hz detection, 200 Hz control, 13.65

Load-bearing premise

The load-bearing premise is that the simulated plant used in the hardware-in-the-loop avoidance tests has similar agility to the real 2.0 kg quadrotor, whose measured physical reaction lag is 0.10–0.23 s in flight; if the simulated plant is more agile, the reported safety margins will not transfer to actual flights.

Editorial extensions

If this is right

With a 4 m sensing range, a quadrotor with acceleration budget 10–200 m/s² can theoretically avoid obstacles at up to 8.8–37 m/s relative speed, since d_crit stays below 3.9 m.
Radar-based DOA works without cameras or GPUs in darkness and smoke, with detection consistency in smoke within the same error spread as light and dark conditions.
A Raspberry Pi 4B can sustain the full perception-to-command chain at 13.65 ms average latency, meeting real-time constraints.
Avoidance rates exceed 80% for objects larger than 0.1 m beyond 1 m range, with closest approach distances above the 0.8 m safety radius in most conditions.
Late detection of very small non-metallic objects (≤0.1 m) and CPU spikes on the Pi can push the trigger distance below d_crit and cause collisions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The derived d_crit formula suggests a simple design rule for any radar-based DOA: if the sensor can detect at range R, the maximum safe relative speed is roughly (R - d_evade)/t_end, which could be used to cap speed or widen margins in cluttered environments.
Because the stationary hardware-in-the-loop experiments apply evasive acceleration to a simulated plant, the reported 89% avoidance and closest-approach distances may overstate real flight safety; the 20 tethered flights with only the large ball provide weaker evidence for small-ball and smoke scenarios.
The system's direct velocity measurement from radar removes the multi-frame velocity-estimation latency that burdens camera and lidar pipelines, so the same architecture could generalize to other fast-moving objects or to multi-obstacle avoidance with a global planner.
The 14 ms latency budget implies that faster object speeds or heavier aircraft (longer t_evade) require either a larger detection range or a shorter t_end; a testable extension is to measure avoidance rate versus object speed and compare with the d_crit prediction.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: the paper deserves a referee, but the referee should push on the simulation backbone of the headline avoidance claim.

The genuinely new content is the integrated system: a 100 Hz mmWave radar detection/tracking pipeline feeding a CBF avoidance controller, running on a Pi4B, with code and a 390-throw dataset. The perception results are MOCAP-anchored (except smoke), the latency breakdown is directly measured (13.65 ms end-to-end on the Pi4B), and the closed-form bounds in Eq. (2) are a real derivation from kinematics, not a fitted curve. The paper also does the right thing in reporting failure conditions.

The soft spots are specific and fixable. Eq. (1) is printed as sqrt(2aR - a*t_end), which is dimensionally wrong and doesn't match Table I; it should be sqrt(2aR) - a*t_end. The headline statistics are computed after Z-score outlier removal with the cutoff stated, but the number of removed samples is never reported, so the 0.15/0.93/0.87 m errors are hard to interpret. Smoke has no ground truth; the std/MAD give dispersion, not accuracy, though the authors honestly say so. And the largest issue, correctly flagged by the stress-test: the 89% avoidance rate and the ds values in Table IV come from a hardware-in-the-loop setup where the quadrotor is physically stationary and the evasive acceleration is applied to a simulated plant that is never described. The real platform shows 0.06–0.23 s reaction lag in flight, so the simulated safety margins may be optimistic. The flight validation itself is limited to hover and 1 m/s ego motion with one object.

None of this kills the central concept—radar as a viable primary DOA sensor on small UAVs without a GPU—and the perception and latency evidence is real. But the system-level safety claim needs a disclosed plant model and preferably a flight test with a faster ego velocity.

Who should read it: people working on sense-and-avoid for small UAVs and on radar perception. I'd bring it to a reading group with the caveats. Recommendation: send to peer review; a serious referee can sort out the simulation/plant disclosure and the typo in a revision.

Referee Report

4 major / 5 minor

Summary. The paper presents a mmWave RADAR-based perception-and-control pipeline for quadrotor dynamic obstacle avoidance. The authors derive a simple latency/spatial bound relating sensing range, relative speed, and control delay, and use it to argue for a minimum sensing range of about 4 m. They implement a lightweight detection/tracking stack (DBSCAN clustering, SAFE-IMM tracker) and a DR-ACBF avoidance controller, and validate it in two settings: stationary hardware-in-the-loop experiments with 390 ball throws in light, dark, and smoke, and 20 onboard flights with a Raspberry Pi 4B. Headline claims include position errors below 0.15 m / 0.93 m / 0.87 m in x / y / z, a 14 ms end-to-end latency, and an overall average of 89% avoidance at obstacle speeds above 3.1 m/s.

Significance. If the result holds, the paper would make a useful contribution: it demonstrates a low-cost, GPU-free, onboard RADAR-based perception and control loop that works in darkness and smoke, provides a simple design rule for sensing range, and ships code and a 390-throw dataset. The latency analysis, while elementary, is presented clearly enough to be reused. The main caveat is that the central safety metrics—89% avoidance and the closest-approach distances d_s in the stationary experiments—are not supported as stated: the avoidance rate is only a trigger rate, and the d_s values appear to come from an unspecified simulated plant while the UAV is physically stationary. These issues undermine the abstract's headline claims and need to be fixed before the paper can be judged as a reliable demonstration of fast DOA.

major comments (4)

[§V-A.2, Table IV] The closest-approach distances d_s and the 'physical reaction delay' t_r in the stationary experiments cannot be measured while the UAV is physically stationary, yet the manuscript does not describe the plant model used to compute the evasive motion. The mass, acceleration limit, actuation delay, and inner-loop dynamics of the simulated plant are absent. Since the abstract's '89% avoidance' and the reported d_s values (minimum 1.07 m) rest on this setup, the paper must either specify and validate the HiL plant model against the flight data (e.g., Table V) or clearly label these metrics as simulated. As written, the safety margins cannot be transferred to the real 2.0 kg quadrotor.
[§V-A, Table III] The 'avoidance rate' is defined as the fraction of throws that trigger an avoidance command, not the fraction of throws that result in a collision-free pass. The abstract and conclusion report 'overall average of 89% avoidance' without this qualification, which conflates command generation with actual collision avoidance. Please rename this metric (e.g., 'trigger rate') and report actual collision-avoidance success or closest-approach distances from flight tests as the primary safety metric.
[§V-D.1, Eq. (2), Table V] The critical-distance calculation omits the measured physical reaction delay t_r. For forward flight, Eq. (2) with t_evade=sqrt(2Rsum/a) and t_end=13.65 ms gives d_crit≈2.53 m at v_rel≈7.43 m/s. Adding the measured t_r≈0.06 s from Table V raises d_crit to ≈2.97 m, which is essentially equal to the dark-condition trigger distance d_t=2.94 m. This contradicts the stated 'robust safety margin.' Clarify whether t_r is included in t_evade, and reconcile with the τ=1.5 s reaction latency used in the DR-CVaR trigger.
[§III, Eq. (1)] Equation (1) as printed, vmax = max(0, sqrt(2aR - a·t_end)), is dimensionally inconsistent and does not reproduce Table I. The table values match vmax = max(0, sqrt(2aR) - a·t_end). Since Eq. (1) is the basis for the sensing-range analysis, please correct the typo and ensure the derivation in the text uses the dimensionally correct form.

minor comments (5)

[§II and Abstract] The novelty claim 'first mmWave RADAR-based perception-and-control system for fast onboard DOA' is undercut by the discussion of [18], which already demonstrated onboard FMCW radar obstacle avoidance on a MAV. Please sharpen the distinction (e.g., 'first at >50 Hz' or 'first for fast dynamic obstacle avoidance') and remove the contradictory statement in §II that 'no prior work presents RADAR-based onboard perception to perform DOA in UAVs.'
[§V-A.1, Table II] The position-error statistics are computed after removing samples with absolute modified Z-score greater than 3.5. Please report the number of removed samples and the raw statistics without outlier rejection, since the abstract's 'less than 0.15 m, 0.93 m, and 0.87 m' phrasing omits this condition.
[§V-A.1, Smoke rows] The smoke-condition rows in Table II report dispersion, not absolute error, because MOCAP ground truth was unavailable. The abstract's 'similar spread for 90 experiments in smoke' is acceptable but should be explicitly caveated as a variability bound, not an accuracy result.
[§V, parameters] The DR-CVaR reaction latency is set to τ=1.5 s with Δ=0.1 s. This is two orders of magnitude larger than the measured t_r and is not used in Eq. (2). Please clarify what τ represents and why this value is appropriate.
[§V-C, Table VI] The reported 13.65 ms end-to-end latency is a sum of mean component latencies on the Pi4B. Since the failure analysis mentions CPU spikes up to 25 ms, please also report worst-case or percentile latencies.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the kinematic bounds are self-contained and the component self-citations are not load-bearing.

full rationale

The paper's central derivation chain consists of Eq. (1) for v_max and Eq. (2) for d_crit, both computed from stated constant-acceleration kinematics with explicit inputs R, a, t_end, and R_sum. No fitted parameter is renamed as a prediction; the assumed latencies (t_det=10 ms, t_avoid=5 ms) and adopted values from [6] are external assumptions, not outputs of the claimed result. The empirical d_crit comparison in Section V-D uses measured t_end=13.65 ms and independently measured trigger distances d_t; the trigger distances are not constructed from d_crit by an equation identity. The avoidance-rate metric is defined as trigger rate rather than physical avoidance, and the stationary HiL setup leaves the simulated plant model for d_s unstated; these are external-validity and reporting concerns, not circular reductions of the claim to its inputs. The self-citations to the authors' prior detection, tracking, and DR-ACBF work are reuse of components, not a self-citation chain that forces the conclusion. No uniqueness theorem is imported, and no ansatz is smuggled in via citation as a substitute for derivation.

Assumptions & free parameters 10 free parameters · 6 assumptions · 0 invented entities

The analytical bounds in Section III need no fitted constants: given R, a, t_end, and Rsum they are closed-form kinematics, up to the Eq. (1) typo. The experimental claims instead depend on numerous hand-chosen parameters: the DR-ACBF risk budgets, SMD gains, Wasserstein radius, clearance, reaction-latency margin, the per-ball z-offset, and an unreported Z-score trimming fraction. The system introduces no new physical entities; it assembles the authors' radar detector [9], tracker [15], and controller [14].

free parameters (10)

DR-ACBF total risk budget alpha_total = 0.15
Hand-chosen total risk budget for the DR-CVaR trigger (Section V).
DR-CVaR risk sensitivity alpha = 0.01
Per-obstacle risk bound feeding the avoidance trigger (Section V).
Wasserstein radius epsilon_wass = 0.05
Ambiguity radius in the DR-ACBF, set to ensure numerical robustness (Section V).
SMD differentiator gains (gamma, L0, L1, L2) = (1.5, 4.0, 3.0, 2.0)
Parameterizes the obstacle acceleration envelope estimation (Section V).
SMD regularization coefficient alpha_smd = 1e-2
Regularization for the acceleration differentiator (Section V).
Lateral clearance d_cl = 1.0 m
Required lateral offset for obstacle clearance in the avoidance controller (Section V).
Reaction latency margin tau = 1.5 s
Used to predict object state for the DR-CVaR trigger, accounting for unknown actuation delay; the physical measured tr is 0.02 to 0.23 s, far below tau (Section V).
Z-score outlier cutoff = 3.5 (absolute modified Z-score)
Samples beyond this are removed before computing every reported error statistic; the removed fraction is never stated (Section V-A).
Per-ball z-axis truth offset = one ball diameter plus marker cap radius
Ad hoc correction assuming RADAR reflects from the ball's bottom surface while MOCAP tracks a top cap; directly reduces the reported z error (Section V-A).
CFAR threshold and DBSCAN (eps, min points) = 15 dB; 0.5; 2 points
Tuned for sparse low-RCS pointclouds and inherited from the authors' detector [9] (Section IV-A, IV-B).

assumptions (6)

domain assumption Constant relative velocity over the reaction and evasion horizon: dcrit = v_rel*(t_end + t_evade)
Section III, Eq. (2). The ball's speed is treated as constant during detection and the whole lateral evasion; thrown balls accelerate through flight, and the tracker's velocity estimate inherits the large y/z errors of Table II.
domain assumption The UAV instantly realizes the commanded lateral acceleration a after t_end
Section III. Actuator dynamics are excluded from the bound; the measured physical reaction delay tr of 0.02 to 0.23 s (Tables IV and V) is not part of the formula.
domain assumption MOCAP output is exact ground truth for ball and UAV poses
Section V-A. Every error and ds metric uses MOCAP, with no error budget given for a thrown, spinning, non-metallic ball or for occlusion during motion.
domain assumption The CFAR/DBSCAN detector of [9] transfers to this setup with the claimed RCS sensitivity
Section IV-B. Detection thresholds, sensitivity to 0.001 m^2 RCS, and detection-rate behavior are inherited from the authors' prior paper without independent replication.
domain assumption The DR-ACBF safety guarantees of [14] hold under the CVaR ambiguity sets and Cantelli bounds used here
Section IV-D. The controller is the authors' prior RAL result; this letter validates it only indirectly through trigger distances and closest approaches.
ad hoc to paper RADAR reflects from the bottom surface of the ball, so the z-truth is offset by one diameter plus cap radius
Section V-A. The z-correction mechanically lowers the reported z error; the assumed reflection point is not independently verified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of RADAR Perception for Dynamic Obstacle Avoidance onboard small-scale Quadrotor UAVs." pith.science (2026). https://pith.science/paper/UVEDM34G

@misc{pith2026260801855,
  author       = {Pith},
  title        = {Pith review of: RADAR Perception for Dynamic Obstacle Avoidance onboard small-scale Quadrotor UAVs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UVEDM34G}},
  note         = {Machine review of arXiv:2608.01855}
}

read the original abstract

Fast dynamic obstacle avoidance (DOA) on uncrewed aerial vehicles (UAVs) demands not only low-latency control and actuation but also reliable perception with sufficient sensing range for accurate obstacle detection and speed estimation. This letter presents, to the best of our knowledge, the first mmWave RADAR-based perception-and-control system for fast onboard DOA. We derive and analyze latency and spatial bounds that relate sensing range, relative speed, and control delay, yielding sufficient conditions for successful avoidance. Our system adopts a lightweight tracker based on interacting multiple models and a controller based on control-barrier functions that directly outputs evasive accelerations. It achieves position errors of less than 0.15 m, 0.93 m, and 0.87 m in x, y, and z directions for 300 experiments with three different object sizes and varying visibility (light and dark), and a similar spread for 90 experiments in smoke. An onboard implementation on a Raspberry Pi 4B demonstrates real-time feasibility with an end-to-end sensing-to-command latency of approximately 14 ms. Code and the full dataset of 390 throws are available (https://tinyurl.com/radardoagit).

Figures

Figures reproduced from arXiv: 2608.01855 by the authors.

**Figure 2.** Experimental setup of our quadrotor with mmWave RADAR sensor. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** UAV control loop of proposed DOA system. The mmWave RADAR [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Three scene conditions for the hardware-in-the-loop experiment as [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 1 canonical work pages

[1]

Unmanned aerial vehicles (uavs): Collision avoidance systems and approaches,

J. N. Yasin, S. A. Mohamed, M.-H. Haghbayan, J. Heikkonen, H. Ten- hunen, and J. Plosila, “Unmanned aerial vehicles (uavs): Collision avoidance systems and approaches,”IEEE access, vol. 8, pp. 105 139– 105 155, 2020

2020
[2]

Perception and avoidance of multiple small fast moving objects for quadrotors with only low-cost rgbd camera,

M. Lu, H. Chen, and P. Lu, “Perception and avoidance of multiple small fast moving objects for quadrotors with only low-cost rgbd camera,”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 11 657–11 664, 2022

2022
[3]

Avoiding dynamic small obstacles with onboard sensing and computation on aerial robots,

F. Kong, W. Xu, Y . Cai, and F. Zhang, “Avoiding dynamic small obstacles with onboard sensing and computation on aerial robots,” IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 7869–7876, 2021

2021
[4]

Robust and efficient depth-based obstacle avoidance for autonomous miniaturized uavs,

H. M ¨uller, V . Niculescu, T. Polonelli, M. Magno, and L. Benini, “Robust and efficient depth-based obstacle avoidance for autonomous miniaturized uavs,”IEEE Transactions on Robotics, vol. 39, no. 6, pp. 4935–4951, 2023

2023
[5]

Dynamic obstacle avoidance for quadrotors with event cameras,

D. Falanga, K. Kleber, and D. Scaramuzza, “Dynamic obstacle avoidance for quadrotors with event cameras,”Science Robotics, vol. 5, no. 40, p. eaaz9712, 2020. [Online]. Available: https: //www.science.org/doi/abs/10.1126/scirobotics.aaz9712

work page doi:10.1126/scirobotics.aaz9712 2020
[6]

How fast is too fast? the role of perception latency in high-speed sense and avoid,

D. Falanga, S. Kim, and D. Scaramuzza, “How fast is too fast? the role of perception latency in high-speed sense and avoid,”IEEE Robotics and Automation Letters, vol. 4, no. 2, pp. 1884–1891, 2019

2019
[7]

Rapid dynamic obstacle avoidance for uavs enhanced by dvs and neuromorphic computing,

S. Wang, S. Yu, T. Liang, Y . Shi, Y . Ma, and P. Ren, “Rapid dynamic obstacle avoidance for uavs enhanced by dvs and neuromorphic computing,” inProc. IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 16 155–16 161

2025
[8]

Dynamic perception-enhanced motion planning and control for uavs flights in challenging dynamic envi- ronments,

L. Liu, J. Xu, and H. Zhang, “Dynamic perception-enhanced motion planning and control for uavs flights in challenging dynamic envi- ronments,” inProc. IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 2794–2800

2025

Show all 22 references

[9]

Evaluating millimeter-wave radar for resource-efficient drone perception,

D. Mandaokar and B. Rinner, “Evaluating millimeter-wave radar for resource-efficient drone perception,” inProc. International Conference on Distributed Computing in Smart Systems and the Internet of Things (DCOSS-IoT), 2025, pp. 352–359

2025
[10]

Autonomous obstacle avoidance for uav based on point cloud,

N. Shen, J. Cao, M. Zipp, and W. Stork, “Autonomous obstacle avoidance for uav based on point cloud,” inProc. International Conference on Unmanned Aircraft Systems (ICUAS), 2022, pp. 1580– 1585

2022
[11]

Pursuing drones with drones using millimeter wave radar,

S. Dogru and L. Marques, “Pursuing drones with drones using millimeter wave radar,”IEEE Robotics and Automation Letters, vol. 5, no. 3, pp. 4156–4163, 2020

2020
[12]

Batmobility: Towards flying without seeing for autonomous drones,

E. Sie, Z. Liu, and D. Vasisht, “Batmobility: Towards flying without seeing for autonomous drones,” inProc. Annual International Conference on Mobile Computing and Networking. ACM, 2023. [Online]. Available: https://doi.org/10.1145/3570361.3592532

2023
[13]

A deep-learning method based on the multistage fusion of radar and camera in uav obstacle avoidance,

X. Wang, X. Wang, Z. Zhou, and Y . Song, “A deep-learning method based on the multistage fusion of radar and camera in uav obstacle avoidance,”IEEE Transactions on Aerospace and Electronic Systems, vol. 60, no. 5, pp. 6734–6751, 2024

2024
[14]

Distributionally robust acceleration control barrier filter for efficient uav obstacle avoidance,

D. Mandaokar and B. Rinner, “Distributionally robust acceleration control barrier filter for efficient uav obstacle avoidance,”IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5914–5921, 2026

2026
[15]

Safe-imm: Robust and lightweight radar-based object tracking on mobile platforms,

——, “Safe-imm: Robust and lightweight radar-based object tracking on mobile platforms,” inProc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2026, pp. 20 282– 20 286

2026
[16]

Dynamic obstacle avoidance for uavs using a fast trajectory planning approach,

H. Chen, P. Lu, and C. Xiao, “Dynamic obstacle avoidance for uavs using a fast trajectory planning approach,” inProc. IEEE International Conference on Robotics and Biomimetics (ROBIO), 2019, pp. 1459– 1464

2019
[17]

Fapp: Fast and adaptive perception and planning for uavs in dynamic cluttered environments,

M. Lu, X. Fan, H. Chen, and P. Lu, “Fapp: Fast and adaptive perception and planning for uavs in dynamic cluttered environments,”IEEE Transactions on Robotics, vol. 41, pp. 871–886, 2025

2025
[18]

Obstacle avoidance onboard mavs using a fmcw radar,

N. Wessendorp, R. Dinaux, J. Dupeyroux, and G. C. H. E. de Croon, “Obstacle avoidance onboard mavs using a fmcw radar,” in Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021, pp. 117–122

2021
[19]

Onboard dynamic-object detection and tracking for autonomous robot naviga- tion with rgb-d camera,

Z. Xu, X. Zhan, Y . Xiu, C. Suzuki, and K. Shimada, “Onboard dynamic-object detection and tracking for autonomous robot naviga- tion with rgb-d camera,”IEEE Robotics and Automation Letters, vol. 9, no. 1, pp. 651–658, 2024

2024
[20]

Control barrier function based quadratic programs for safety critical systems,

A. D. Ames, X. Xu, J. W. Grizzle, and P. Tabuada, “Control barrier function based quadratic programs for safety critical systems,”IEEE Transactions on Automatic Control, vol. 62, no. 8, pp. 3861–3876, 2017

2017
[21]

Cubedn: Real-time drone detection in 3d space from dual mmwave radar cubes,

Y . Fang, F. Shi, X. Wei, Q. Chen, K. Chetty, and S. Julier, “Cubedn: Real-time drone detection in 3d space from dual mmwave radar cubes,” inProc. IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025, pp. 3977–3983

2025
[22]

3d-sensing mimo radar for uav formation flight and obstacle avoidance,

B. B. Tierney and C. T. Rodenbeck, “3d-sensing mimo radar for uav formation flight and obstacle avoidance,” inProc. IEEE Radio and Wireless Symposium (RWS), 2019, pp. 1–3

2019

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Unmanned aerial vehicles (uavs): Collision avoidance systems and approaches,

J. N. Yasin, S. A. Mohamed, M.-H. Haghbayan, J. Heikkonen, H. Ten- hunen, and J. Plosila, “Unmanned aerial vehicles (uavs): Collision avoidance systems and approaches,”IEEE access, vol. 8, pp. 105 139– 105 155, 2020

2020

[2] [2]

Perception and avoidance of multiple small fast moving objects for quadrotors with only low-cost rgbd camera,

M. Lu, H. Chen, and P. Lu, “Perception and avoidance of multiple small fast moving objects for quadrotors with only low-cost rgbd camera,”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 11 657–11 664, 2022

2022

[3] [3]

Avoiding dynamic small obstacles with onboard sensing and computation on aerial robots,

F. Kong, W. Xu, Y . Cai, and F. Zhang, “Avoiding dynamic small obstacles with onboard sensing and computation on aerial robots,” IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 7869–7876, 2021

2021

[4] [4]

Robust and efficient depth-based obstacle avoidance for autonomous miniaturized uavs,

H. M ¨uller, V . Niculescu, T. Polonelli, M. Magno, and L. Benini, “Robust and efficient depth-based obstacle avoidance for autonomous miniaturized uavs,”IEEE Transactions on Robotics, vol. 39, no. 6, pp. 4935–4951, 2023

2023

[5] [5]

Dynamic obstacle avoidance for quadrotors with event cameras,

D. Falanga, K. Kleber, and D. Scaramuzza, “Dynamic obstacle avoidance for quadrotors with event cameras,”Science Robotics, vol. 5, no. 40, p. eaaz9712, 2020. [Online]. Available: https: //www.science.org/doi/abs/10.1126/scirobotics.aaz9712

work page doi:10.1126/scirobotics.aaz9712 2020

[6] [6]

How fast is too fast? the role of perception latency in high-speed sense and avoid,

D. Falanga, S. Kim, and D. Scaramuzza, “How fast is too fast? the role of perception latency in high-speed sense and avoid,”IEEE Robotics and Automation Letters, vol. 4, no. 2, pp. 1884–1891, 2019

2019

[7] [7]

Rapid dynamic obstacle avoidance for uavs enhanced by dvs and neuromorphic computing,

S. Wang, S. Yu, T. Liang, Y . Shi, Y . Ma, and P. Ren, “Rapid dynamic obstacle avoidance for uavs enhanced by dvs and neuromorphic computing,” inProc. IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 16 155–16 161

2025

[8] [8]

Dynamic perception-enhanced motion planning and control for uavs flights in challenging dynamic envi- ronments,

L. Liu, J. Xu, and H. Zhang, “Dynamic perception-enhanced motion planning and control for uavs flights in challenging dynamic envi- ronments,” inProc. IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 2794–2800

2025