Flying a quadrotor through a cluttered environment requires not only planning a collision-free reference trajectory based on perceived obstacles, but the reference also needs to be dynamically feasible and within the actuation limits of the vehicle, so that the controller can track it precisely. Existing methods either optimize a smooth polynomial inside a convex corridor, which limits agility, or treat obstacles as soft costs traded against tracking performance. We propose a Nonlinear Model Predictive Planning (NMPP) that imposes perceived obstacles as hard geometric constraints and hands a full-state reference to an obstacle-blind SE(3) controller. Our planner achieves a 58-67 % lower position RMSE than a linear Model Predictive Control trajectory planner and a 41-70 % lower RMSE than a polynomial trajectory planner. It also completes all forest flights with up to 9.5 m/s speed without collisions, and achieves 86 % flight success rate under a more aggressive speed profile where a state-of-the-art planner has only 26 % success rate. The real-world deployment showed reliable execution flying up to 5.5 m/s in an unknown cluttered environment.
Figures & tables
Fig. 1 : Illustrative images from real-world experiments: (a) UAV used in the experiments, (b) flown path visualization in RViz, and (c) top view of the course with the flown path.
Fig. 2 : Whole nmpp system pipeline.
Profile
vxy
axy
jxy
vz
az↑ / ↓
[ ms−1 ]
[ ms−2 ]
[ ms−3 ]
[ ms−1 ]
[ ms−2 ]
slow
1.0
1.0
20
1.0
1.0/1.0
medium
4.0
2.0
40
2.0
1.0/1.0
fast
8.0
4.0
60
4.0
2.0/2.0
super fast
10.0
10.0
120
3.0
3.0/2.0
agile
13.0
15.0
120
3.0
4.0/2.0
TABLE I : The constraint profiles. The snap limit equals the jerk limit in every profile.
Profile
vpeak
NMPP
MPC
Poly
Red.
Red.
(proposed)
(baseline)
(baseline)
vs MPC
vs Poly
slow
1.0–1.2
0.032
0.077
0.055
58.2
41.3
medium
3.9–4.1
0.078
0.238
0.192
67.3
59.5
fast
5.9–7.2
0.122
0.373
0.307
67.4
60.3
super fast
7.3–8.8
0.178
0.497
0.398
64.1
55.2
agile
9.5–10.0
0.186
0.545
0.624
65.9
70.2
TABLE II : Mean 3D rmse ( m ) of a 20m flight in free space for varying constraint profiles. vpeak is the range of peak speeds reached across the three planners in ( ms−1 ) while Reduction denotes the percentage decrease in error relative to each baseline. No standard deviation exceeded 0.045m .
Profile
RMSE [ m ]
vmax [ ms−1 ]
tmean [ ms ]
tp95 [ ms ]
slow
0.041
1.47
5.51
7.88
medium
0.096
4.29
5.86
8.66
fast
0.143
7.08
5.95
9.57
super fast
0.172
9.51
5.74
8.92
agile
0.198
11.38
6.91
9.79
TABLE III : Tracking and computational performance of nmpp in the simulated forest. The rmse is pooled over the ten flights of each profile. tmean and tp95 cover the nmpc update only.
Fig. 3 : nmpp solution time over the ten simulated forest flights under the super fast profile.
Fig. 4 : Predicted trajectories obtained during the ten super fast simulation runs. Obstacles are shown in black with red-dashed safety margins.
Profile
NMPP (proposed)
SUPER
succ. [ % ]
min. clr. [ m ]
succ. [ % ]
min. clr. [ m ]
slow
100
+1.243
100
-0.041
medium
100
+0.800
100
+0.004
fast
100
+0.270
100
+0.183
super fast
100
+0.264
70
+0.158
agile
86
+0.131
26
-0.145
TABLE IV : Success rate and clearance in the forest. Min. clr. is the smallest airframe-to-trunk gap over the successful flights, with the propeller tips reaching 0.231m from the vehicle centre, so a negative value is a propeller strike the flight survived. The agile row is from fifty flights per method.
Nflights
RMSE [ m ]
vmax [ ms−1 ]
tmean [ ms ]
tp95 [ ms ]
10
0.328±0.070
5.55
13.52
20.03
TABLE V : Aggregated performance over the ten real-world flights under the fast profile. The rmse is the mean and standard deviation over the individual flights, vmax denotes maximum achieved flight speed and tmean and tp95 denote the mean and 95th-percentile nmpp computation times, respectively.
Fig. 5 : Real uav trajectories with corresponding speeds.
Autonomous UAV flight through cluttered and partially unknown environments requires reasoning not only about observed obstacles but also about occluded regions that the sensor cannot observe. We present OA-MPPI, an obstacle- and occlusion-aware extension of Model Predictive Path Integral (MPPI) control for quadrotor flight that accounts for potential moving agents emerging from these regions into the vehicle's path. At every planning step, we extract a 3D occlusion boundary from the online occupancy map and use it to model the regions that hidden agents could reach over the prediction horizon. We penalize trajectories that enter these expanding regions within MPPI rollouts generated using nonlinear quadrotor dynamics and accounting for individual rotor thrust limits. We validate the proposed approach in simulation and hardware flight experiments, with the complete pipeline running onboard the vehicle in real time. Results show increased clearance from occlusion boundaries compared to baseline MPPI in both settings, as well as avoidance of an agent emerging from occlusion in simulation.
Vittorio Palladino, Teaya Yang, Ruiqi Zhang +1
High Performance Robotics Laboratory, Department of Mechanical Engineering, University of California, Berkeley, CA 94720, United States
This paper presents a new robust integrated planning and control (IPC) strategy for multirotor uncrewed aerial vehicles. We propose a nonlinear model predictive control (NMPC) formulation that embeds control barrier functions (CBFs) as exponential penalties, improving feasibility while ensuring smooth obstacle avoidance under tight input bounds. The penalty weights provide a practical tuning knob to trade off tracking accuracy against avoidance aggressiveness. We enhance the system robustness by employing a high-gain disturbance observer (HGDO) to estimate and compensate for external disturbances. We also incorporate a Kalman filter (KF) for computationally efficient, real-time prediction of obstacle motion, enabling avoidance of moving obstacles. Comparative studies against both conventional NMPC and NMPC with hard CBF constraints, validated in Gazebo and hardware experiments, demonstrate superior feasibility, safety, and robustness. To the best of our knowledge, this is the first hardware-validated NMPC-CBF IPC framework, offering a practical step toward safe quadrotor deployment in dynamic environments.
Zeinab Shayan, Mohammadreza Izadi, Reza Faieghi
Autonomous Vehicles Laboratory, Department of Aerospace Engineering, Toronto Metropolitan University, Toronto, Canada.
This paper presents an approach to mutual collision avoidance based on Nonlinear Model Predictive Control (NMPC) with time-dependent Reciprocal Velocity Constraints (RVCs). Unlike most existing methods, the proposed approach relies solely on observable information about other robots, eliminating the need for excessive communication. The computationally efficient algorithm for computing RVCs, together with the direct integration of these constraints into the NMPC problem formulation at the controller level, allows the whole pipeline to run at 100 Hz. This high processing rate, combined with modeled nonlinear dynamics of the controlled Uncrewed Aerial Vehicles (UAVs), is a key feature that facilitates the use of the proposed approach for agile UAV flight. The proposed approach was evaluated through extensive simulations emulating real-world conditions in scenarios involving up to 10 UAVs and velocities of up to 25 m/s, and in real-world experiments with accelerations up to 30 m/s2. Comparison with the state of the art shows 31% improvement in terms of flight time reduction in challenging scenarios, while maintaining a collision-free navigation in all trials.
Vit Kratky, Robert Penicka, Parakh M. Gupta +2
Department of Cybernetics, Faculty of Electrical Engineering, Czech Technical University in Prague, Czech Republic