Terrain-Aware Perceptual Planning for Aerial Vehicles in Martian Environments

Jessica Eve Todd, Pedro Roque, Adam Johnson, Joel W. Burdick

Keywords: Perception-Aware Navigation, Model Predictive Path Integral, Semantic Classifier, Terrain Classifier, Active-Vision, Planetary Robotics

TL;DR: Terrain-aware planning and control towards for improved visual-inertial odometry and state-estimation in aerial planetary robotics.

Abstract: Robust state estimation is a cornerstone of successful planetary robotics missions. Visual-inertial odometry (VIO) has emerged as a leading approach in this domain, providing computationally efficient state estimation through the fusion of camera and inertial measurement unit (IMU) data. However, with its reliance on trackable visual features, VIO performance is inherently coupled to the visual quality of the surrounding terrain, which varies significantly across planetary surfaces. To address this challenge, we propose a terrain-aware Model Predictive Path Integral (TA-MPPI) control framework that actively incorporates terrain visual-feature quality into the planning process, guiding the robot towards regions that support reliable state estimation. We evaluate our approach in synthetic worlds and a simulated Mars environment using the PX4-Autopilot software-in-the-loop stack and OpenVINS. Our results demonstrate that terrain-quality-aware planning improves VIO performance compared to terrain-agnostic baselines, highlighting the benefit of incorporating perception awareness into motion planning.

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: 8