He Haodong

He Haodong

Undergraduate at Tongji University & Research Assistant at MARMot Lab, NUS

Heterogeneous Multi-Agent Systems under Communication and Planning Constraints

CBS Protocol Project

Source: Veerapaneni R, Tang A, He H, et al. Conflict-Based Search as a Protocol: A Multi-Agent Motion Planning Protocol for Heterogeneous Agents, Solvers, and Independent Tasks. arXiv:2510.00425, 2025.

Big Picture: Heterogeneous Multi-Agent Systems

The future of autonomy will not be dominated by a single type of robot or a single algorithm. Instead, it will involve teams of heterogeneous agents—aerial drones and ground vehicles, manipulators and mobile bases, learning-driven agents and classical optimizers—working together in complex, uncertain environments.

Heterogeneity appears in three dimensions:

  • Hardware diversity: robots differ in form factors, sensing modalities, and actuation capabilities.
  • Algorithmic diversity: planners span heuristic search, sampling methods, optimization, diffusion models, and reinforcement learning.
  • Operational diversity: agents face varying levels of uncertainty, resource constraints, and communication availability.

To unlock the full potential of such systems, we need general coordination frameworks that:

  1. Abstract away low-level differences, so that agents can cooperate without having to conform to a single planning or control paradigm.
  2. Adapt to uncertainty and partial information, enabling resilience in real-world deployments.
  3. Balance autonomy and cooperation, letting agents contribute individually while still aligning with team-level goals.

My research vision is to build this umbrella of heterogeneous multi-agent autonomy, where different agents can be integrated seamlessly into a shared decision-making protocol.

Focus Project: CBS Protocol

(Joint work with Rishi Veerapaneni)

One concrete step toward this vision is the CBS Protocol, where we extend the well-known Conflict-Based Search (CBS) framework.
🔗 Project Page

The CBS Protocol extends the classic Conflict-Based Search (CBS) framework (Sharon et al., 2015) to act as a general coordination protocol for heterogeneous agents.

  • Key Idea: Require only one API from each agent: the ability to return a collision-free path that satisfies given space-time constraints.
  • Planner-Agnostic: With this interface, agents can be powered by A*, RRT, Direct Collocation, Diffusion Models, or Reinforcement Learning—and CBS can still coordinate them.
  • Centralized Conflict Resolution: A central planner uses CBS to resolve conflicts at the team level, ensuring global collision-free trajectories while leaving local planning details to each agent’s algorithm.

  • Outcome: This protocol demonstrates, for the first time, efficient multi-agent motion planning with algorithmically heterogeneous teams performing independent tasks.

Looking Ahead

The CBS Protocol highlights a promising direction: treating coordination not as a monolithic algorithm, but as a protocol that can flexibly integrate heterogeneous planning modules. Looking forward, I aim to extend these principles toward large-scale multi-robot systems that combine the guarantees of classical planning with the adaptability of learning-based methods, all under a unifying coordination framework.