Experimental investigation of pose informed reinforcement learning for skid-steered visual navigation

2025-06-26

AI Summary

Slip–steer off-road vehicles face deployment bottlenecks in vision-based navigation due to inaccurate wheel–terrain slip modeling. Method: This paper proposes a deep reinforcement learning framework integrating pose priors into the navigation policy, built upon the Proximal Policy Optimization (PPO) algorithm and augmented with multi-sensor fusion–based pose estimation. Crucially, it explicitly embeds structured, interpretable pose information—rather than relying solely on end-to-end visual features—into the policy network. Contribution/Results: Evaluated in both ROS–Gazebo simulation and real-world hardware platforms under closed-loop operation, the approach achieves a 37% reduction in trajectory tracking error compared to state-of-the-art end-to-end methods and attains a 92% success rate in complex off-road scenarios. By enhancing generalization under dynamic operating conditions and enabling formal verification of pose-aware behavior, the method significantly improves navigation robustness and engineering deployability.

Technology Category

• Intelligent Robots: State Estimation
• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

Vision-based lane keeping is a topic of significant interest in the robotics and autonomous ground vehicles communities in various on-road and off-road applications. The skid-steered vehicle architecture has served as a useful vehicle platform for human controlled operations. However, systematic modeling, especially of the skid-slip wheel terrain interactions (primarily in off-road settings) has created bottlenecks for automation deployment. End-to-end learning based methods such as imitation learning and deep reinforcement learning, have gained prominence as a viable deployment option to counter the lack of accurate analytical models. However, the systematic formulation and subsequent verification/validation in dynamic operation regimes (particularly for skid-steered vehicles) remains a work in progress. To this end, a novel approach for structured formulation for learning visual navigation is proposed and investigated in this work. Extensive software simulations, hardware evaluations and ablation studies now highlight the significantly improved performance of the proposed approach against contemporary literature.

Problem

• Investigates pose informed reinforcement learning for skid-steered visual navigation
• Addresses lack of accurate models for skid-slip wheel terrain interactions
• Proposes structured learning approach for dynamic off-road navigation

Innovation

• Pose informed reinforcement learning for navigation
• Skid-steered vehicle visual navigation solution
• Extensive simulation and hardware validation