Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Soft Actor-Critic-based Control Barrier Adaptation for Robust Autonomous Navigation in Unknown Environments

Nicholas Mohammad, Nicola Bezzo

TL;DR

This work addresses the safety-performance trade-off in autonomous navigation by adaptively tuning the ZCBF safety parameter $α$ with a Soft Actor-Critic policy, enabling safe yet non-conservative tracking in unknown cluttered environments. The method integrates with generic high-level planners and low-level controllers and is trained in high-fidelity simulations, with validation in both simulation and real-world experiments using MPC and PD controllers. Key contributions include (i) an open-source CBF implementation with SAC-based online $α$ adaptation for broad compatibility, and (ii) a SAC training pipeline that operates outside OpenAI Gym, allowing training in realistic simulators and safe real-world refinement. The results demonstrate improved safety while maintaining progress toward the goal, highlighting practical potential for robust navigation in unknown environments.

Abstract

Motion planning failures during autonomous navigation often occur when safety constraints are either too conservative, leading to deadlocks, or too liberal, resulting in collisions. To improve robustness, a robot must dynamically adapt its safety constraints to ensure it reaches its goal while balancing safety and performance measures. To this end, we propose a Soft Actor-Critic (SAC)-based policy for adapting Control Barrier Function (CBF) constraint parameters at runtime, ensuring safe yet non-conservative motion. The proposed approach is designed for a general high-level motion planner, low-level controller, and target system model, and is trained in simulation only. Through extensive simulations and physical experiments, we demonstrate that our framework effectively adapts CBF constraints, enabling the robot to reach its final goal without compromising safety.

Soft Actor-Critic-based Control Barrier Adaptation for Robust Autonomous Navigation in Unknown Environments

TL;DR

This work addresses the safety-performance trade-off in autonomous navigation by adaptively tuning the ZCBF safety parameter with a Soft Actor-Critic policy, enabling safe yet non-conservative tracking in unknown cluttered environments. The method integrates with generic high-level planners and low-level controllers and is trained in high-fidelity simulations, with validation in both simulation and real-world experiments using MPC and PD controllers. Key contributions include (i) an open-source CBF implementation with SAC-based online adaptation for broad compatibility, and (ii) a SAC training pipeline that operates outside OpenAI Gym, allowing training in realistic simulators and safe real-world refinement. The results demonstrate improved safety while maintaining progress toward the goal, highlighting practical potential for robust navigation in unknown environments.

Abstract

Motion planning failures during autonomous navigation often occur when safety constraints are either too conservative, leading to deadlocks, or too liberal, resulting in collisions. To improve robustness, a robot must dynamically adapt its safety constraints to ensure it reaches its goal while balancing safety and performance measures. To this end, we propose a Soft Actor-Critic (SAC)-based policy for adapting Control Barrier Function (CBF) constraint parameters at runtime, ensuring safe yet non-conservative motion. The proposed approach is designed for a general high-level motion planner, low-level controller, and target system model, and is trained in simulation only. Through extensive simulations and physical experiments, we demonstrate that our framework effectively adapts CBF constraints, enabling the robot to reach its final goal without compromising safety.

Paper Structure

This paper contains 16 sections, 19 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Control Barrier Function Preliminaries
  4. Problem Formulation
  5. Approach
  6. ZCBF Formulation
  7. Dynamic $\alpha$ Adjustment
  8. SAC for $\alpha$ Adaptation
  9. State and Reward Design:
  10. SAC Training:
  11. Simulations
  12. Case Study 1: Model Predictive Controller
  13. Case Study 2: PD Controller
  14. Physical Experiments
  15. Conclusions and Future Work
  16. ...and 1 more sections

Figures (5)

  • Figure 1: (Top) Traditional motion planning framework fails to avoid obstacles when navigating a narrow corridor. (Bottom) Proposed framework adapts CBF constraints in order to reach the final goal while ensuring safety.
  • Figure 2: Block diagram for SAC training and online deployment.
  • Figure 3: (a) Illustration of SAC state $\bm{s}_t$. (b) Base navigation pipeline crashing while tracking a generated trajectory $\bm{r}(\cdot)$. (c) Using CBF safety filter with $\alpha=.5$ still results in a collision due to infeasible constraints. (d)-(e) Full approach navigating the environment while adapting $\alpha$ as needed.
  • Figure 4: Simulation results for MPC. (a) Test world 5 in Gazebo. (b) Crash with fixed $\alpha=.5$. (c), (d) Success with the full approach. (e) Success rate difference between full approach and $\alpha=.5$. (f) $\alpha(t)$ plot for full approach.
  • Figure 5: (Left) Jackal navigating through a cluttered environment with the full approach and (Top Right) crashing without CBF filter. (Bottom Right) shows $\alpha$ adaptation using our full approach.