Papers
Topics
Authors
Recent
Search
2000 character limit reached

Quadrupedal Robot Skateboard Mounting via Reverse Curriculum Learning

Published 10 May 2025 in cs.RO, cs.AI, and math.OC | (2505.06561v1)

Abstract: The aim of this work is to enable quadrupedal robots to mount skateboards using Reverse Curriculum Reinforcement Learning. Although prior work has demonstrated skateboarding for quadrupeds that are already positioned on the board, the initial mounting phase still poses a significant challenge. A goal-oriented methodology was adopted, beginning with the terminal phases of the task and progressively increasing the complexity of the problem definition to approximate the desired objective. The learning process was initiated with the skateboard rigidly fixed within the global coordinate frame and the robot positioned directly above it. Through gradual relaxation of these initial conditions, the learned policy demonstrated robustness to variations in skateboard position and orientation, ultimately exhibiting a successful transfer to scenarios involving a mobile skateboard. The code, trained models, and reproducible examples are available at the following link: https://github.com/dancher00/quadruped-skateboard-mounting

Summary

Quadrupedal Robot Skateboard Mounting via Reverse Curriculum Learning

The paper titled "Quadrupedal Robot Skateboard Mounting via Reverse Curriculum Learning" explores a methodology to empower quadrupedal robots to autonomously mount skateboards, tackling a complex problem often overlooked in robotic skateboarding scenarios. The research focuses on an intricate part of robot mobility—the mounting phase of a skateboard, which has not been sufficiently studied in existing literature.

The authors propose employing Reverse Curriculum Reinforcement Learning (RCRL) to conduct this task. This approach initiates a learning process from goal states, progressively expanding to incorporate more challenging scenarios. By fixing the skateboard in the primary stages of training, they provide a stable and predictable environment for the robot to focus on learning the mounting behavior. Through the gradual relaxation of constraints, the robots learn to adapt to varying skateboard positions, handing the complexities of an unfixed skateboard in final training stages.

Methodology

Control Approach

The research posits employing Reinforcement Learning (RL) over Model Predictive Control (MPC) for addressing the skateboard mounting task. While MPC is known for its sophistication in planning motions, RL offers robust real-time environmental adaptability without the computational rigidity associated with planning. Among RL algorithms, Proximal Policy Optimization (PPO) was selected due to its stability and efficacy in continuous control tasks. The paper also integrates Reverse Curriculum Learning, a technique well-suited for sparse-reward problems, where traditional RL struggles with the exploratory aspect.

Experimentation Design

The study utilizes an extended robotic simulation environment from RobotLab-Isaac for undertaking trials. A realistic skateboard model was designed with accurate physical properties for simulation purposes. Observations in the experiment capture the robot's kinematics, skateboard configuration, and foot contact with the skateboard surface, enhancing adaptation to real-world scenarios.

Reverse Curriculum Learning Implementation

The Reverse Curriculum Learning method employed significantly enhances exploration efficacy in complex sparse-reward tasks. The robot was initially positioned above a fixed skateboard, gradually diversifying initial conditions and external perturbations as training progressed. The strategy was proven practical, allowing the quadrupedal robot to achieve mounting objectives autonomously.

Results

As training outcomes demonstrate, the quadrupedal robot effectively learns skateboard mounting from various starting positions with consistent robustness and reliability. The experimental evidence indicates the utility of the proposed method to achieve a successful transfer from simulation to the practical application in potentially real-world scenarios without exhibiting degradation in locomotion capabilities.

Conclusions and Future Directions

The Reverse Curriculum Reinforcement Learning methodology illustrates considerable promise for developing autonomous mounting capabilities in quadrupedal robots on skateboards, bridging existing gaps from traditional locomotion patterns to riding environments.

Future work could integrate elements like steering control, combining mounting and riding skills into a seamless controller, and validating the approach through real-world experimentation. The paper opens pathways to designing more competent robotic skateboard riders, further enhancing the versatility and adaptability of quadrupedal robots in interacting with everyday dynamic environments.

Paper to Video (Beta)

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Collections

Sign up for free to add this paper to one or more collections.