Guided Reinforcement Learning for Omnidirectional 3D Jumping in Quadruped Robots

2025-07-22

AI Summary

Quadrupedal robots face significant challenges in omnidirectional 3D jumping, including low sample efficiency, unpredictable motion behavior, and poor robustness. To address these issues, we propose a physics-guided reinforcement learning (RL) framework: (1) trajectory planning is explicitly parameterized using Bézier curves combined with a constant-acceleration linear motion model to encode interpretable physical priors; and (2) a kinematic consistency–based reward shaping mechanism is introduced to enable efficient and stable policy training. This design substantially reduces sample complexity while enhancing policy safety and interpretability. Extensive simulations and real-world experiments demonstrate that our method achieves higher success rates and superior environmental adaptability across diverse jumping tasks—outperforming state-of-the-art end-to-end RL and trajectory optimization approaches. The framework establishes a new paradigm for high-dynamic locomotion control in embodied agents.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Jumping poses a significant challenge for quadruped robots, despite being crucial for many operational scenarios. While optimisation methods exist for controlling such motions, they are often time-consuming and demand extensive knowledge of robot and terrain parameters, making them less robust in real-world scenarios. Reinforcement learning (RL) is emerging as a viable alternative, yet conventional end-to-end approaches lack efficiency in terms of sample complexity, requiring extensive training in simulations, and predictability of the final motion, which makes it difficult to certify the safety of the final motion. To overcome these limitations, this paper introduces a novel guided reinforcement learning approach that leverages physical intuition for efficient and explainable jumping, by combining Bézier curves with a Uniformly Accelerated Rectilinear Motion (UARM) model. Extensive simulation and experimental results clearly demonstrate the advantages of our approach over existing alternatives.

Problem

• Overcoming inefficiency in quadruped robot jumping control
• Reducing sample complexity in reinforcement learning for jumping
• Ensuring safety and predictability in robot jumping motions

Innovation

• Combines Bézier curves with UARM model
• Uses guided reinforcement learning for jumping
• Enhances efficiency and motion explainability