Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer-Based Model Predictive Path Integral Control

Shrenik Zinage, Vrushabh Zinage, Efstathios Bakolas

TL;DR

TransformerMPPI addresses the inefficiency of sampling-based MPPI by using a transformer to generate an informed initial mean control sequence. The model is trained on historical MPPI trajectories via teacher forcing, enabling real-time predictions that steer MPPI sampling toward promising regions of the control space. Across 2D navigation with obstacles and autonomous racing, TransformerMPPI delivers lower average costs, improved sample efficiency, and faster computational speed compared to conventional MPPI. This approach preserves compatibility with existing MPPI variants and holds promise for real-time robotic control in dynamic environments.

Abstract

This paper presents a novel approach to improve the Model Predictive Path Integral (MPPI) control by using a transformer to initialize the mean control sequence. Traditional MPPI methods often struggle with sample efficiency and computational costs due to suboptimal initial rollouts. We propose TransformerMPPI, which uses a transformer trained on historical control data to generate informed initial mean control sequences. TransformerMPPI combines the strengths of the attention mechanism in transformers and sampling-based control, leading to improved computational performance and sample efficiency. The ability of the transformer to capture long-horizon patterns in optimal control sequences allows TransformerMPPI to start from a more informed control sequence, reducing the number of samples required, and accelerating convergence to optimal control sequence. We evaluate our method on various control tasks, including avoidance of collisions in a 2D environment and autonomous racing in the presence of static and dynamic obstacles. Numerical simulations demonstrate that TransformerMPPI consistently outperforms traditional MPPI algorithms in terms of overall average cost, sample efficiency, and computational speed in the presence of static and dynamic obstacles.

Transformer-Based Model Predictive Path Integral Control

TL;DR

TransformerMPPI addresses the inefficiency of sampling-based MPPI by using a transformer to generate an informed initial mean control sequence. The model is trained on historical MPPI trajectories via teacher forcing, enabling real-time predictions that steer MPPI sampling toward promising regions of the control space. Across 2D navigation with obstacles and autonomous racing, TransformerMPPI delivers lower average costs, improved sample efficiency, and faster computational speed compared to conventional MPPI. This approach preserves compatibility with existing MPPI variants and holds promise for real-time robotic control in dynamic environments.

Abstract

This paper presents a novel approach to improve the Model Predictive Path Integral (MPPI) control by using a transformer to initialize the mean control sequence. Traditional MPPI methods often struggle with sample efficiency and computational costs due to suboptimal initial rollouts. We propose TransformerMPPI, which uses a transformer trained on historical control data to generate informed initial mean control sequences. TransformerMPPI combines the strengths of the attention mechanism in transformers and sampling-based control, leading to improved computational performance and sample efficiency. The ability of the transformer to capture long-horizon patterns in optimal control sequences allows TransformerMPPI to start from a more informed control sequence, reducing the number of samples required, and accelerating convergence to optimal control sequence. We evaluate our method on various control tasks, including avoidance of collisions in a 2D environment and autonomous racing in the presence of static and dynamic obstacles. Numerical simulations demonstrate that TransformerMPPI consistently outperforms traditional MPPI algorithms in terms of overall average cost, sample efficiency, and computational speed in the presence of static and dynamic obstacles.

Paper Structure

This paper contains 12 sections, 26 equations, 14 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Formulation and Preliminaries
  3. Problem Formulation
  4. MPPI Algorithm
  5. Proposed approach
  6. Transformer architecture for informed mean control sequence initialization
  7. TransformerMPPI Algorithm
  8. Data collection and training
  9. Results
  10. Navigation 2D
  11. Autonomous racing
  12. Conclusions

Figures (14)

  • Figure 1: MPPI
  • Figure 2: TransformerMPPI
  • Figure 4: TransformerMPPI
  • Figure 6: Transformer
  • Figure 7: Encoder block
  • ...and 9 more figures