Team Spirit! (Ongoing)

Dynamic Obstacle Avoidance and Loco-Manipulation planning for a quadruped

Introduction

Quadruped robots face significant challenges in dynamic obstacle avoidance and loco-manipulation due to rapidly changing environments and this has been a prime research topic till now. To address this, we propose the implementation of a PRM planner that supports loco-manipulation tasks by planning in the z-direction as well. The proposed kinodynamic planner has been tested out in the simulation framework called QUAD-SDK.

Motivation

The current implementation in the QUAD-SDK stack for the quadruped is the RRT-Connect algorithm that plans in x and y directions and faces difficulty in navigation through changing its height as shown in the video below.

Current shortcoming of the 2-axis RRT-Connect Planner.

PRM Implementation:

Key Features:

  • PRM roadmap precomputed with nodes representing feasible configurations.
  • Dynamic obstacle roadmap updates using environment feedback.
  • Foothold selection and balance constraints incorporated.
  • Adapted z-axis height for loco-manipulation tasks, ensuring stable trajectories.
Implementation of PRM Planner

Results

Performance Metrics: Path quality: Improved by ~30% compared to RRT-Connect. Planning time: Reduced by ~20% under high obstacle density.

EKF Framework and the hardware setup.

Visual Demonstration

Below is a video demonstrating the quadruped navigating a dynamic environment using PRM.

Watch the EKF in action, predicting and correcting cloth state in real-time.

Challenges and Future Work

Challenges

  • Efficient roadmap updates in highly dynamic scenarios.
  • Balancing computational load with real-time requirements.

Future Work

  • Explore hybrid planners combining PRM and reactive methods.
  • Use reinforcement learning to optimize roadmap generation.

References