Path Integral Control with Rollout Clustering and Dynamic Obstacles (2403.18066v1)

Abstract: Model Predictive Path Integral (MPPI) control has proven to be a powerful tool for the control of uncertain systems (such as systems subject to disturbances and systems with unmodeled dynamics). One important limitation of the baseline MPPI algorithm is that it does not utilize simulated trajectories to their fullest extent. For one, it assumes that the average of all trajectories weighted by their performance index will be a safe trajectory. In this paper, multiple examples are shown where the previous assumption does not hold, and a trajectory clustering technique is presented that reduces the chances of the weighted average crossing in an unsafe region. Secondly, MPPI does not account for dynamic obstacles, so the authors put forward a novel cost function that accounts for dynamic obstacles without adding significant computation time to the overall algorithm. The novel contributions proposed in this paper were evaluated with extensive simulations to demonstrate improvements upon the state-of-the-art MPPI techniques.

• The paper introduces rollout clustering in MPPI to segregate trajectories and enhance safe path selection in uncertain environments.
• It employs a DBSCAN-based clustering method and a novel cost function to effectively handle dynamic obstacles while maintaining efficiency.
• Experiments demonstrate reduced collisions and consistent computation time, validating the method's performance in both static and dynamic scenarios.

Path Integral Control with Rollout Clustering and Dynamic Obstacles

Introduction

The paper introduces enhancements to the Model Predictive Path Integral (MPPI) control method, specifically targeting its application in environments characterized by uncertainty and dynamic obstacles. The authors identify key limitations in MPPI's utilization of trajectory simulations and its inability to effectively handle dynamic obstacles. This paper proposes novel techniques in trajectory clustering and cost function design to overcome these limitations.

Methodology

Model Predictive Path Integral Control

MPPI is a control strategy that optimizes a sequence of control inputs based on numerous simulated trajectories. These trajectories are perturbed using random variables to account for system uncertainties. The optimal control sequence minimizes a cost function composed of running and terminal costs. Existing MPPI approaches faced challenges in scenarios with significant disturbance noise, where naive averaging of perturbed trajectories often led to unsafe paths.

Rollout Clustering

To address the shortcomings in trajectory utilization, the authors propose a clustering method based on the DBSCAN algorithm. This clustering segregates trajectories into groups with similar cost function behaviors, reducing the probability that the weighted average of all trajectories traverses unsafe regions. A truncated Gaussian is used for importance sampling within each cluster, ensuring that the sampling distribution remains valid.

Dynamic Obstacles Handling

For dynamic environments, a novel cost function is introduced, parametrized by the approximate trajectories of moving obstacles. This method distinctly models dynamic obstacle trajectories, allowing for efficient computation by avoiding the need to resample obstacle states for each perturbation. By precomputing likely obstacle paths, the system can anticipate and navigate around dynamic obstacles without excessive computation.

Results

Static Obstacles

In scenarios with static obstacles, the proposed clustering method significantly reduced the incidence of collision and failures compared to standard MPPI and related advanced methods. The evaluation demonstrated a consistent computation time with enhanced safety, showing evidence of improved path integrity under perturbations.

Dynamic Obstacles

In dynamic environments, the DC-MPPI (Dynamic and Clustered MPPI) method notably reduced collision occurrences. By incorporating dynamic obstacle trajectories into the cost function, the algorithm maintained obstacle avoidance with a modest increase in computation time. The results indicate a substantial advantage in terms of safety and effectiveness when dealing with environments containing moving entities.

Conclusion

This paper provides two key advancements for MPPI: trajectory clustering and dynamic obstacle handling, each addressing core limitations observed in traditional implementations. The clustering ensures that risky paths are segregated, enhancing safety and robustness, while the novel cost function efficiently integrates dynamic obstacle data, improving real-time planning capabilities. These improvements maintain computational efficiency, showcasing the practicality of integrating these techniques into existing MPPI frameworks. Future research could explore the integration of these methods with various sensor inputs and different environmental scenarios to further assess performance.