Occupancy-Aware Reasoning for Safe Quadrotor Navigation: Perception-Aware MPPI

Yifan Zhai, Rudolf Reiter, Davide Scaramuzza

Keywords: Quadrotor navigation, occupancy mapping, model predictive path integral, perception-aware control, safe autonomy, foundation model

TL;DR: Control-level navigation through partially known environments through semantic maps.

Abstract:

Safe robot autonomy in unstructured environments demands geometric sensing, and, importantly,
robots must reason about what regions of the environment are traversable,
occupied, or simply unknown.
We present PAMPPI Perception-Aware Model Predictive Path Integral Control,
a real-time quadrotor controller that tightly couples occupancy-aware map reasoning with
sampling-based optimal control.
PAMPPI maintains a three-state occupancy grid
(free, occupied, unknown) and introduces a novel perception cost that
steers optimized trajectories toward unknown frontiers aligned with the goal,
enabling exploration-driven navigation without external planners or reference
trajectories.
Running at 50Hz, PAMPPI performs on par with the state-of-the-art
safety-assured planner SUPER across challenging, cluttered scenes, while
achieving up to 34% lower energy consumption.
We further demonstrate that PAMPPI serves as a safe and robust action policy
for navigation foundation models, compensating for their lack of 3-D geometric
awareness.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 11