Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Compositional Diffusion Models for Powered Descent Trajectory Generation with Flexible Constraints

Julia Briden, Yilun Du, Enrico M. Zucchelli, Richard Linares

TL;DR

This work introduces TrajDiffuser, a compositional diffusion-based flexible and concurrent trajectory generator for 6 degrees of freedom powered descent guidance that enables efficient optimizer initialization, increasing its robustness and speed.

Abstract

This work introduces TrajDiffuser, a compositional diffusion-based flexible and concurrent trajectory generator for 6 degrees of freedom powered descent guidance. TrajDiffuser is a statistical model that learns the multi-modal distributions of a dataset of simulated optimal trajectories, each subject to only one or few constraints that may vary for different trajectories. During inference, the trajectory is generated simultaneously over time, providing stable long-horizon planning, and constraints can be composed together, increasing the model's generalizability and decreasing the training data required. The generated trajectory is then used to initialize an optimizer, increasing its robustness and speed.

Compositional Diffusion Models for Powered Descent Trajectory Generation with Flexible Constraints

TL;DR

This work introduces TrajDiffuser, a compositional diffusion-based flexible and concurrent trajectory generator for 6 degrees of freedom powered descent guidance that enables efficient optimizer initialization, increasing its robustness and speed.

Abstract

This work introduces TrajDiffuser, a compositional diffusion-based flexible and concurrent trajectory generator for 6 degrees of freedom powered descent guidance. TrajDiffuser is a statistical model that learns the multi-modal distributions of a dataset of simulated optimal trajectories, each subject to only one or few constraints that may vary for different trajectories. During inference, the trajectory is generated simultaneously over time, providing stable long-horizon planning, and constraints can be composed together, increasing the model's generalizability and decreasing the training data required. The generated trajectory is then used to initialize an optimizer, increasing its robustness and speed.
Paper Structure (24 sections, 61 equations, 16 figures, 4 tables)

This paper contains 24 sections, 61 equations, 16 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Paper Organization
  3. Background
  4. Diffusion Models
  5. Training
  6. Sampling
  7. Diffusion Models as Energy-Based Models
  8. Compositional Diffusion
  9. Reverse Diffusion Sampling for Compositional Diffusion
  10. Six Degrees-of-Freedom Powered Descent Guidance
  11. State-Triggered Constraints
  12. Velocity-Triggered Angle of Attack Constraint
  13. Experiments
  14. Minimum-Time Six Degree-of-Freedom Powered Descent Guidance Diffusion Model
  15. Diffusion Model Architecture
  16. ...and 9 more sections

Figures (16)

  • Figure 1: The forward and reverse processes in Gaussian diffusion.
  • Figure 2: Diffusion model reverse sampled trajectories from TrajDiffuser. The scattered points show the trajectory for every 200 steps in the backward process, increasing opacity as the trajectory converges. Each color indicates a separate trajectory.
  • Figure 3: Generated Trajectories - Position. The lefthand plot shows 50 trajectories generated by the diffusion model, and the righthand plot shows 50 trajectories uniformly sampled by the trajectory optimization framework.
  • Figure 4: Mass Over Time. The lefthand plot shows 50 mass values over time generated by the diffusion model, and the righthand plot shows 50 mass values over time uniformly sampled by the trajectory optimization framework.
  • Figure 5: Velocity Components Over Time. The lefthand plot shows 50 velocity values over time generated by the diffusion model, and the righthand plot shows 50 velocity values over time uniformly sampled by the trajectory optimization framework.
  • ...and 11 more figures