OA-MPPI: Occlusion-Aware Model Predictive Path Integral Control for UAV Flight

Summary

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.

Abstract Summary

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.

Key Points

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.

Preliminaries

MPPI draws a large number of candidate action sequences via Monte Carlo sampling and combines them into an improved sequence based on their cost. For a single rollout $k$, the cost takes the general receding-horizon form where $V(\bm{x}_{H})$ is the terminal cost, $\ell(\bm{x}_{j},\bm{u}_{j})$ the stage cost applied at every step of the horizon $H$, and $\bm{x}_{j}$ the state of rollout $k$ at stage $j$.

Methodology

We adopt the frame convention of [ 16 ] : ${B}$ is the body-fixed frame, with $\bm{u}_{3}^{B}$ perpendicular to the propeller plane, and ${E}$ the earth-fixed inertial frame, with $\bm{u}_{3}^{E}$ opposite to gravity.

The state is $\bm{p}:=\bm{s}_{BT}\in\real^{3}$ (position relative to a fixed earth point $T$), $\bm{q}:=\bm{q}^{BE}\in\mathbb{S}\mathbb{O}(3)$ (attitude), $\bm{v}:=\bm{v}_{BE}\in\real^{3}$ (velocity), and $\bm{\omega}:=\bm{\omega}^{BE}\in\real^{3}$ (body rate), so $\bm{x}=[\bm{p},\bm{q},\bm{v},\bm{\omega}]$.

Methodology — OA-MPPI: Occlusion-Aware Model Predictive Path Integral Control for UAV Flight — Fig. 4: Simulation of an agent emerging from occlusion. The drone collides with the agent under baseline MPPI (gray), while OA-MPPI (blue) avoids the collision. The agent is shown as a purple sphere and speed $0.4$ m/s. Top: $t=2.8$ s; bottom: $t=3.9$ s.

Results

To validate the effectiveness of our approach, we evaluate it both in simulation and on real hardware. In simulation, a ray-casting sensor model [ 19 ] attached to the vehicle body samples the scene, and the resulting scan is consumed directly by ROG-Map [ 18 ] . In flight experiments, everything runs onboard a Jetson Orin Nano with $8$ GB of RAM.

In both cases, rollouts are evaluated in parallel via OpenMP across the available threads. Both the simulation tests and the real flight experiments use the same MPPI parameters, listed in Table I .

Results — OA-MPPI: Occlusion-Aware Model Predictive Path Integral Control for UAV Flight — Fig. 5: Top: simulation setup for flight over an occluding box. Bottom: side view in the $xz$-plane at $t=2.1$ s. OA-MPPI (blue) takes a higher path over the box than baseline MPPI (gray), increasing clearance from the occlusion boundary.

Sources and Demonstrations

Figures are reproduced from Palladino et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.