SEB-Naver: A SE(2)-based Local Navigation Framework for Car-like Robots on Uneven Terrain

TL;DR

SEB-Naver addresses the challenge of autonomous car-like robot navigation on uneven terrain by coupling a GPU-accelerated $SE(2)$ local mapping framework with a differential-flatness-inspired trajectory optimizer that accounts for terrain-induced $SE(3)$ kinematics. The system integrates LIO-based localization, real-time $SE(2)$ elevation/risk mapping, and an optimization-based planner that uses flat outputs and an intermediate variable to avoid nonholonomic singularities, achieving efficient planning and accurate tracking. Extensive simulations and real-world tests demonstrate real-time performance and improved trajectory quality, including the ability to reverse on challenging terrain. The approach offers a practical, scalable solution for real-time navigation on uneven terrain with car-like robots, though it relies on accurate terrain modeling and state estimation and currently does not address slip or deformable terrain dynamics.

Abstract

Autonomous navigation of car-like robots on uneven terrain poses unique challenges compared to flat terrain, particularly in traversability assessment and terrain-associated kinematic modelling for motion planning. This paper introduces SEB-Naver, a novel SE(2)-based local navigation framework designed to overcome these challenges. First, we propose an efficient traversability assessment method for SE(2) grids, leveraging GPU parallel computing to enable real-time updates and maintenance of local maps. Second, inspired by differential flatness, we present an optimization-based trajectory planning method that integrates terrain-associated kinematic models, significantly improving both planning efficiency and trajectory quality. Finally, we unify these components into SEB-Naver, achieving real-time terrain assessment and trajectory optimization. Extensive simulations and real-world experiments demonstrate the effectiveness and efficiency of our approach. The code is at https://github.com/ZJU-FAST-Lab/seb_naver.

Figures (7)

• Figure 1: Figure (a) illustrates that different robot poses with different pitch and roll angles may be projected onto the same grid cell in $\mathbb{R}^2$. Figure (b) illustrates that the throttle required for the robot to achieve the same acceleration $a_x$ is different due to the presence of gravity.
• Figure 2: Overview of the framework. LiDAR-Inertial Odometry (LIO) receives information from IMU and point clouds from LiDAR to calculate pose. Then, the local mapping module acquires the point cloud and poses information to update the elevation and evaluate the traversability in $SE(2)$ space. The resulting risk maps are used by the local planner. After path search and trajectory optimization, the local planner outputs a trajectory in $SE(2)$ space, which is given to the MPC controller for generating final control commands to send to the robot.
• Figure 3: Local mapping process in $SE(2)$ space.
• Figure 4: One test of the SEB-Naver on the pump track.
• Figure 5: Processing time comparison of $SE(2)$ grids.