Robodog: Quadruped Locomotion Environments in Isaac Sim/Lab

HARE Lab, UC Santa Cruz — September 2024 – June 2025


Overview

As an undergraduate researcher at the HARE Lab, I contributed to the development of simulation infrastructure for reinforcement learning-based quadruped locomotion on the Unitree B1 robot. The work focused on building out the Isaac Sim/Lab environment pipeline — designing terrain conditions, configuring simulation tooling, and setting up experiment scaffolding to support future RL policy training.

While I graduated before the lab reached full policy training, I was able to design and implement a variety of terrain environments in Isaac Sim 4.5.0, including the rough terrain shown below.


Terrain Design

Rough terrain mesh in Isaac Sim 4.5.0
Rough terrain mesh generated in Isaac Sim 4.5.0 for quadruped locomotion training.

Terrain conditions were designed to challenge locomotion policies across a range of difficulty levels. Diverse terrain variations included rough uneven ground, slopes, steps, and mixed surfaces — all parameterized to enable curriculum learning during RL training.


Contributions


Tools & Stack


← Back to Portfolio