Efficient Motion Planning for Manipulators with Control Barrier Function-Induced Neural Controller

TL;DR

This paper addresses real-time, safe motion planning for high-DOF robotic manipulators in cluttered environments by introducing a neural CBF-induced steering controller (CBF-INC) that integrates with RRT. It proposes two observation-driven CBF-INF variants, sCBF-INF (state-based) and oCBF-INF (LiDAR-based), and leverages a QP-based safety layer (CBF-INC) to steer planning toward feasible regions, balancing safety and goal-reaching better than hand-crafted CBFs. Empirical results show substantial improvements in success rate and exploration efficiency in simulation for 4-DoF and 7-DoF arms and demonstrate hardware viability, including robust performance under partial observability and dynamic environments. The approach enables real-time, safe exploration by combining the strengths of CBFs for safety with RRT’s long-horizon planning, guided by neural estimators trained offline on diverse observations. Practical impact includes safer, more efficient motion planning for complex manipulators in clutter, with applicability to real-world robotic systems.

Abstract

Sampling-based motion planning methods for manipulators in crowded environments often suffer from expensive collision checking and high sampling complexity, which make them difficult to use in real time. To address this issue, we propose a new generalizable control barrier function (CBF)-based steering controller to reduce the number of samples needed in a sampling-based motion planner RRT. Our method combines the strength of CBF for real-time collision-avoidance control and RRT for long-horizon motion planning, by using CBF-induced neural controller (CBF-INC) to generate control signals that steer the system towards sampled configurations by RRT. CBF-INC is learned as Neural Networks and has two variants handling different inputs, respectively: state (signed distance) input and point-cloud input from LiDAR. In the latter case, we also study two different settings: fully and partially observed environmental information. Compared to manually crafted CBF which suffers from over-approximating robot geometry, CBF-INC can balance safety and goal-reaching better without being over-conservative. Given state-based input, our neural CBF-induced neural controller-enhanced RRT (CBF-INC-RRT) can increase the success rate by 14% while reducing the number of nodes explored by 30%, compared with vanilla RRT on hard test cases. Given LiDAR input where vanilla RRT is not directly applicable, we demonstrate that our CBF-INC-RRT can improve the success rate by 10%, compared with planning with other steering controllers. Our project page with supplementary material is at https://mit-realm.github.io/CBF-INC-RRT-website/.

Figures (7)

• Figure 1: An example of a Franka Emika Panda robot planning motions using our proposed framework. We train a CBF-induced neural controller and integrate it into the steer function in sampling-based motion planning. The controller is used for safe exploration and steers the edge to collision-free space without being over-conservative.
• Figure 2: Illustration of observations. Left: sCBF-INF takes the signed distance to the nearest obstacle as observation. Middle: For fully-observable environments, oCBF-INF uses a point cloud sampled uniformly on the obstacles as observation. Right: For partially-observable environments, oCBF-INF observes the point cloud from a mounted LiDAR sensor.
• Figure 3: The overall neural network architecture. Left: The architecture of the sCBF-INF. Right: The architecture of the oCBF-INF.
• Figure 4: Motion planning experiments under LiDAR-based setting.
• Figure 5: Left: Slices of oCBF-INF value for Panda, obtained by sweeping across two joints, with all other joint states and obstacle positions held constant. Right: Learning curves of constraint satisfaction rate on Panda.