Teddy-Liao/walk-these-ways-go2

What it solves

This project enables Sim-to-Real locomotion for the Unitree Go2 quadruped robot. It adapts the "walk-these-ways" starter kit (originally designed for the Go1) to the Go2's newer architecture, specifically updating the SDK interfaces to move away from UDP-based communication to the unitree_sdk2.

How it works

The system uses a reinforcement learning pipeline where a policy is trained in a simulated environment using Isaac Gym and then deployed to the physical robot. To bridge the gap between the high-level Python policy and the robot's low-level hardware, the project implements a communication layer using Lightweight Communications and Marshalling (LCM) and a custom C++ bridge (lcm_position_go2) that interfaces with the unitree_sdk2.

Who it’s for

Robotics researchers and developers working with Unitree Go2 robots who want to implement and test learned locomotion policies using a Sim-to-Real workflow.

Highlights

• SDK Adaptation: Specifically modified to support the unitree_sdk2 architecture for the Go2 robot.
• Sim-to-Real Pipeline: Includes tools for training in Isaac Gym and deploying the resulting policy to hardware.
• Flexible Deployment: Supports deployment via an external PC/laptop via Ethernet or directly on the onboard Nvidia Jetson Orin.
• Pretrained Models: Provides a pretrained model for gait-conditioned agility to accelerate getting started.