Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Agent Formation Navigation Using Diffusion-Based Trajectory Generation

Hieu Do Quang, Chien Truong-Quoc, Quoc Van Tran

TL;DR

This work develops a diffusion-based planner for leader–follower multi-agent formation control in cluttered environments by predicting the midpoint trajectory of two leaders treated as a rigid bar in the plane.A diffusion policy generates long action sequences for the midpoint, while followers maintain the formation via a distributed distance-based controller based on relative local coordinates, enabling smooth, low-error motion with multimodal planning capabilities.Key contributions include adapting DDPMs for planar formation planning with obstacle-aware conditioning, a CNN-based diffusion backbone with FiLM conditioning, and a comprehensive experimental comparison against MPPI and path-aware controllers.Results show the diffusion approach can produce efficient, smooth trajectories and near-zero formation errors in many scenarios, but performance degrades in narrow or unseen obstacle configurations, pointing to the need for broader training data and potentially transformer-based backbones for improved long-range planning.

Abstract

This paper introduces a diffusion-based planner for leader--follower formation control in cluttered environments. The diffusion policy is used to generate the trajectory of the midpoint of two leaders as a rigid bar in the plane, thereby defining their desired motion paths in a planar formation. While the followers track the leaders and form desired foramtion geometry using a distance-constrained formation controller based only on the relative positions in followers' local coordinates. The proposed approach produces smooth motions and low tracking errors, with most failures occurring in narrow obstacle-free space, or obstacle configurations that are not in the training data set. Simulation results demonstrate the potential of diffusion models for reliable multi-agent formation planning.

Multi-Agent Formation Navigation Using Diffusion-Based Trajectory Generation

TL;DR

This work develops a diffusion-based planner for leader–follower multi-agent formation control in cluttered environments by predicting the midpoint trajectory of two leaders treated as a rigid bar in the plane.A diffusion policy generates long action sequences for the midpoint, while followers maintain the formation via a distributed distance-based controller based on relative local coordinates, enabling smooth, low-error motion with multimodal planning capabilities.Key contributions include adapting DDPMs for planar formation planning with obstacle-aware conditioning, a CNN-based diffusion backbone with FiLM conditioning, and a comprehensive experimental comparison against MPPI and path-aware controllers.Results show the diffusion approach can produce efficient, smooth trajectories and near-zero formation errors in many scenarios, but performance degrades in narrow or unseen obstacle configurations, pointing to the need for broader training data and potentially transformer-based backbones for improved long-range planning.

Abstract

This paper introduces a diffusion-based planner for leader--follower formation control in cluttered environments. The diffusion policy is used to generate the trajectory of the midpoint of two leaders as a rigid bar in the plane, thereby defining their desired motion paths in a planar formation. While the followers track the leaders and form desired foramtion geometry using a distance-constrained formation controller based only on the relative positions in followers' local coordinates. The proposed approach produces smooth motions and low tracking errors, with most failures occurring in narrow obstacle-free space, or obstacle configurations that are not in the training data set. Simulation results demonstrate the potential of diffusion models for reliable multi-agent formation planning.
Paper Structure (14 sections, 20 equations, 3 figures, 3 tables, 2 algorithms)

This paper contains 14 sections, 20 equations, 3 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Methodology
  3. Graph Theory
  4. Problem Formulation
  5. Denoising Diffusion Probabilistic Models (DDPM)
  6. Diffusion for path planning
  7. Path planning
  8. Neural Network Architecture
  9. Training and Sampling process
  10. DISTANCE-BASED FORMATION TRACKING CONTROL
  11. EXPERIMENTS
  12. Experimental setup
  13. Experiment results
  14. CONCLUSION AND FUTURE WORK

Figures (3)

  • Figure 1: Overview of the proposed diffusion-based trajectory generation and formation tracking framework.
  • Figure 2: Formation trajectories in different environments.
  • Figure 3: Representative failure cases