Decentralized Lifelong Path Planning for Multiple Ackerman Car-Like Robots

TL;DR

This work tackles lifelong multi-robot path planning for non-holonomic car-like robots in continuous spaces by introducing two complementary algorithms: PBCR, a decentralized PIBT-inspired method that uses motion primitives and a count-based exploration heuristic to avoid deadlocks, and ECCR, a centralized enhancement of CL-CBS that leverages focal hybrid A* planning for suboptimal yet efficient trajectory generation. PBCR prioritizes scalability and fast reactivity, yielding high success rates in dense scenarios, while ECCR delivers shorter, higher-quality trajectories with fewer high-level expansions. The methods are validated through extensive simulations, lifelong task experiments, and real-robot demonstrations, showing practical viability and clear trade-offs between throughput, trajectory length, and planning time. Overall, the paper advances practical lifelong path planning for car-like robots by combining discrete-search-inspired strategies with continuous-domain motion primitives and targeted heuristics, enabling robust, scalable operation in static and lifelong contexts.

Abstract

Path planning for multiple non-holonomic robots in continuous domains constitutes a difficult robotics challenge with many applications. Despite significant recent progress on the topic, computationally efficient and high-quality solutions are lacking, especially in lifelong settings where robots must continuously take on new tasks. In this work, we make it possible to extend key ideas enabling state-of-the-art (SOTA) methods for multi-robot planning in discrete domains to the motion planning of multiple Ackerman (car-like) robots in lifelong settings, yielding high-performance centralized and decentralized planners. Our planners compute trajectories that allow the robots to reach precise $SE(2)$ goal poses. The effectiveness of our methods is thoroughly evaluated and confirmed using both simulation and real-world experiments.

Figures (8)

• Figure 1: A simulated example on a $60\times 30$ map with 10 obstacles and robots. The collision-free trajectories for the car-like robots are shown.
• Figure 2: (a) Ackermann steering kinematic model. (b) The predefined motion primitives. We have at most 8 motion primitives, which are forward max-left (FL), forward straight (FS), forward max- right (FR), backward max-left (BL), backward straight (BS), backward max-right (BR), wait, and greedy motion primitive (GM)
• Figure 3: An illustrative case highlighting the potential occurrence of deadlocks due to the exclusion of the count-based heuristic is as follows: In this scenario, two robots, labeled as robot 1 and robot 2, travel in opposite directions along a linear path. When robot 2 yields to robot 1 in PBCR, a predicament arises. In this situation, the motion primitives FS, FR, and FL lead to collisions, compelling robot 2 to consistently opt for the motion primitive BS. This choice is driven by BL and BR, involving additional turning penalties. A dynamic shift occurs in priorities upon robot 1's successful arrival at its designated goal state. Robot 1 is assigned a lower priority than robot 2 and consequently yields to the latter. Similarly, robot 1 consistently selects BS using the greedy strategy. Consequently, an unending cycle emerges, entangling robot 1 and robot 2 in an indefinite sequence of movements
• Figure 4: Evaluation results on a $100\times 100$ empty map for a varying number of robots.
• Figure 5: Evaluation data on a $100\times 100$ map with $50$ randomly placed obstacles for a varying number of robots.