Guided Reinforcement Learning for Omnidirectional 3D Jumping in Quadruped Robots

2025/07/22 by Riccardo Bussola, Bussola, Riccardo, Michele Focchi +7 · 2 citations
Engineering · #Control and Dynamics of Mobile Robots #FOS: Computer and information sciences #FOS: Electrical engineering #Modular Robots and Swarm Intelligence #Robotic Locomotion and Control #Robotics (cs.RO) #Systems and Control (eess.SY) #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.2507.16481

openalex publication_date 2025/07/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

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.

Citations

• Quadruped-Frog: Rapid Online Optimization of Continuous Quadruped Jumping
• Cafe-Mpc: A Cascaded-Fidelity Model Predictive Control Framework with Tuning-Free Whole-Body Control
• Curriculum-Based Reinforcement Learning for Quadrupedal Jumping: A Reference-free Design
• ManyQuadrupeds: Learning a Single Locomotion Policy for Diverse Quadruped Robots
• Two-Stage Learning of Highly Dynamic Motions with Rigid and Articulated Soft Quadrupeds
• ANYmal Parkour: Learning Agile Navigation for Quadrupedal Robots
• Locosim: an Open-Source Cross-Platform Robotics Framework
• Continuous Versatile Jumping Using Learned Action Residuals
• GenLoco: Generalized Locomotion Controllers for Quadrupedal Robots
• An Optimal Motion Planning Framework for Quadruped Jumping
• Agile Maneuvers in Legged Robots: a Predictive Control Approach
• Learning Torque Control for Quadrupedal Locomotion
• Rapid and Reliable Quadruped Motion Planning with Omnidirectional Jumping
• Contact-timing and Trajectory Optimization for 3D Jumping on Quadruped Robots
• Deep Reinforcement Learning Versus Evolution Strategies: A Comparative Survey
• Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement\n Learning
• Model-free Reinforcement Learning for Robust Locomotion using Demonstrations from Trajectory Optimization
• Online Trajectory Optimization for Dynamic Aerial Motions of a Quadruped Robot
• Robust Quadruped Jumping via Deep Reinforcement Learning
• Learning Agile Robotic Locomotion Skills by Imitating Animals
• Training in Task Space to Speed Up and Guide Reinforcement Learning
• Learning agile and dynamic motor skills for legged robots
• Clipped Action Policy Gradient
• Proximal Policy Optimization Algorithms
• Continuous control with deep reinforcement learning

Cited by

• Can SSD-Mamba2 Unlock Reinforcement Learning for End-to-End Motion Control?
• LocoMamba: Vision-Driven Locomotion via End-to-End Deep Reinforcement Learning with Mamba

Related

• Explosive Jumping with Rigid and Articulated Soft Quadrupeds via Example Guided Reinforcement Learning
• Curriculum-Based Reinforcement Learning for Quadrupedal Jumping: A Reference-free Design
• Towards Quadrupedal Jumping and Walking for Dynamic Locomotion using Reinforcement Learning
• Continuous Versatile Jumping Using Learned Action Residuals
• Integrating Trajectory Optimization and Reinforcement Learning for Quadrupedal Jumping with Terrain-Adaptive Landing