EB-MBD: Emerging-Barrier Model-Based Diffusion for Safe Trajectory Optimization in Highly Constrained Environments
Raghav Mishra, Ian Manchester
AI summary
Problem
Standard Model-Based Diffusion algorithms suffer catastrophic performance degradation in highly constrained environments because Monte Carlo score estimation becomes dominated by infeasible "dead" samples, making reliable trajectory optimization difficult.
Approach
The method augments Model-Based Diffusion with time-varying log barrier functions that start relaxed and progressively tighten, guiding the sampling process away from constraint boundaries while maintaining viable samples without expensive projection operations.
Key results
- Demonstrated MBD's catastrophic degradation in highly constrained spaces due to poor score estimation
- Proposed Emerging Barrier MBD with time-varying barriers for guaranteed feasibility and higher-quality solutions
- Analyzed sampling statistics and derived theoretical bounds to guide barrier parameter scheduling
- Achieved lower-cost solutions than MBD and orders-of-magnitude faster computation than projection-based methods
Why it matters
Enables reliable and computationally efficient constrained trajectory optimization for robotics applications that struggle with non-smooth dynamics or lack differentiable models.
Abstract
We propose enforcing constraints on Model-Based Diffusion by introducing emerging barrier functions inspired by interior point methods. We demonstrate that the standard Model-Based Diffusion algorithm can lead to catastrophic performance degradation in highly constrained environments, even on simple 2D systems due to sample inefficiency in the Monte Carlo approximation of the score function. We introduce Emerging-Barrier Model-Based Diffusion (EB-MBD) which uses progressively introduced barrier constraints to avoid these problems, significantly improving solution quality, without expensive projection operations such as projections. We analyze the sampling liveliness of samples at each iteration to inform barrier parameter scheduling choice. We demonstrate results for 2D collision avoidance and a 3D underwater manipulator system and show that our method achieves lower cost solutions than Model-Based Diffusion, and requires orders of magnitude less computation time than projection based methods.