Conflict-Based Search as a Protocol: A Multi-Agent Motion Planning Protocol for Heterogeneous Agents, Solvers, and Independent Tasks

TL;DR

The paper tackles the challenge of coordinating heterogeneous multi-agent systems where each agent runs potentially proprietary single-agent planners. It formalizes Algorithmically Heterogeneous MAMP (AH-MAMP) and proposes CBS as a Protocol that coordinates diverse plan() APIs through space-time constraints, enabling collision-free multi-agent motion planning for independent tasks. The authors demonstrate broad applicability by coordinating solvers across heuristic search, sampling, optimization, diffusion, and reinforcement learning, and by handling tasks beyond start-goal planning. Experimental results on diverse 2D maps show that CBS can integrate heterogeneous solvers, with performance tied to the underlying single-agent solvers and enhanced by experience reuse, suggesting practical potential for real-world heterogeneous robotics applications.

Abstract

Imagine the future construction site, hospital, office, or even sophisticated household with dozens of robots bought from different manufacturers. How can we enable these different systems to effectively move in a shared environment, given that each robot may have its own independent motion planning system? This work shows how we can get efficient collision-free movements between algorithmically heterogeneous agents by using Conflict-Based Search (Sharon et al. 2015) as a protocol. At its core, the CBS Protocol requires one specific single-agent motion planning API; finding a collision-free path that satisfies certain space-time constraints. Given such an API, CBS uses a central planner to find collision-free paths - independent of how the API is implemented. We show how this protocol enables multi-agent motion planning for a heterogeneous team of agents completing independent tasks with a variety of single-agent planners including: Heuristic Search (e.g., A*), Sampling Based Search (e.g., RRT), Optimization (e.g., Direct Collocation), Diffusion, and Reinforcement Learning.

Figures (5)

• Figure 1: We depict using the CBS Protocol (left) and its internal process (right). Left: Given a Multi-Agent Motion Planning (MAMP) problem, each agent defines a plan() API that satisfies the CBS Protocol requirements (blue dashed box). This is input into the protocol (which has no other knowledge except for collision detection) which uses the APIs and outputs a collision-free solution. Right: The CBS Protocol starts with generating a root node by querying each agent's plan() API without constraints. It detects collisions between agents (e.g., agents A and C collide spatially at point $p_0$ at time $t_0$) and generates two child CT nodes. Each node is created by adding a constraint (colored X) that the agent needs to avoid $(p_0,t_0)$ and re-querying the agent's plan() API with the additional constraint. CBS repeats this process of detecting collisions, adding constraints, and querying APIs with additional constraints until a collision-free solution is found.
• Figure 2: An example of the CBS Protocol with 6 heterogeneous agents with different solvers, dynamics, and tasks. Constraints for the solution are plotted as colored x's.
• Figure 3: We show results on MAMP start-goal problems with different solvers and categorize them by the number of conflicts in the root node (a proxy for problem difficulty). The top plot shows the number of CT nodes that the CBS Protocol generates for successful (blue) and unsuccessful (red) instances while the bottom plot shows the overall success rate.
• Figure A1: Examples of successful instances of the CBS Protocol with algorithmically heterogeneous agents on IND2 Maps. Each legend denotes the task, solver, and dynamics for each agent. Note that some agents also have different footprint shapes/sizes. X's denote space-time constraints required for the solution found by CBS, with the time depicted by the color bar.
• Figure A2: Examples of successful instances of the CBS Protocol with algorithmically heterogeneous agents on a map with random obstacles and an empty map. Each legend denotes the task, solver, and dynamics for each agent. Note that some agents also have different footprint shapes/sizes. X's denote space-time constraints required for the solution found by CBS, with the time depicted by the color bar.