Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TSPDiffuser: Diffusion Models as Learned Samplers for Traveling Salesperson Path Planning Problems

Ryo Yonetani

TL;DR

TSPDiffuser is presented, a novel data-driven path planner for traveling salesperson path planning problems (TSPPPs) in environments rich with obstacles that enables efficient and accurate estimation of travel costs between destinations, effectively addressing the primary computational challenge in solving TSPPPs.

Abstract

This paper presents TSPDiffuser, a novel data-driven path planner for traveling salesperson path planning problems (TSPPPs) in environments rich with obstacles. Given a set of destinations within obstacle maps, our objective is to efficiently find the shortest possible collision-free path that visits all the destinations. In TSPDiffuser, we train a diffusion model on a large collection of TSPPP instances and their respective solutions to generate plausible paths for unseen problem instances. The model can then be employed as a learned sampler to construct a roadmap that contains potential solutions with a small number of nodes and edges. This approach enables efficient and accurate estimation of travel costs between destinations, effectively addressing the primary computational challenge in solving TSPPPs. Experimental evaluations with diverse synthetic and real-world indoor/outdoor environments demonstrate the effectiveness of TSPDiffuser over existing methods in terms of the trade-off between solution quality and computational time requirements.

TSPDiffuser: Diffusion Models as Learned Samplers for Traveling Salesperson Path Planning Problems

TL;DR

TSPDiffuser is presented, a novel data-driven path planner for traveling salesperson path planning problems (TSPPPs) in environments rich with obstacles that enables efficient and accurate estimation of travel costs between destinations, effectively addressing the primary computational challenge in solving TSPPPs.

Abstract

This paper presents TSPDiffuser, a novel data-driven path planner for traveling salesperson path planning problems (TSPPPs) in environments rich with obstacles. Given a set of destinations within obstacle maps, our objective is to efficiently find the shortest possible collision-free path that visits all the destinations. In TSPDiffuser, we train a diffusion model on a large collection of TSPPP instances and their respective solutions to generate plausible paths for unseen problem instances. The model can then be employed as a learned sampler to construct a roadmap that contains potential solutions with a small number of nodes and edges. This approach enables efficient and accurate estimation of travel costs between destinations, effectively addressing the primary computational challenge in solving TSPPPs. Experimental evaluations with diverse synthetic and real-world indoor/outdoor environments demonstrate the effectiveness of TSPDiffuser over existing methods in terms of the trade-off between solution quality and computational time requirements.
Paper Structure (25 sections, 3 equations, 5 figures, 1 algorithm)

This paper contains 25 sections, 3 equations, 5 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Path Planning and Data-Driven Planners
  4. Traveling Salesperson Problem (TSP)
  5. Diffusion Models
  6. Preliminaries
  7. Problem Definition
  8. Denoising Diffusion Probabilistic Models (DDPMs)
  9. TSPDiffuser
  10. Model Architecture
  11. Constructing Roadmaps using Learned Diffusion Models
  12. Solving TSPPP
  13. Experiments
  14. Planning Environments
  15. Evaluation Setups
  16. ...and 10 more sections

Figures (5)

  • Figure 1: Solving TSPPPs with TSPDiffuser (right). We leverage a diffusion model as a learned sampler to construct a roadmap for traveling salesperson path planning problems (TSPPPs). Our approach significantly improves the balance between solution quality and efficiency compared to existing sampling-based path planners such as probabilistic roadmaps (PRM; left)
  • Figure 2: Solution Overview. (a) Problem instance. (b) Generating multiple paths (each shown in different colors) using a learned diffusion model. (c) Connecting paths to construct a roadmap. (d) Perform path planning and solving TSP on the roadmap to obtain a solution path.
  • Figure 3: Model Architecture. We represent a TSPPP instance using binary and distance images of obstacles and destinations, which are stacked and encoded into a unified feature map. Noisy position data, concatenated with denoising timestep information and the feature vector extracted from the feature map, are fed into a diffusion model to reconstruct the original clean path.
  • Figure 4: Quantitative Results. (a1-a5, d1-d3) SPL scores and required computation times. (b) Specific time breakdown for TSPDiffuser and PRM. (c) Ablation study.
  • Figure 5: Qualitative Results. Roadmaps and solution paths are visualized with gray and black lines (if solved). Obstacle regions are colored in blue, and destinations are marked with red circles.