Wheeled Lab
Wheeled Lab

Wheeled Lab

Introduction

Wheeled Lab is a full-stack integration of Isaac Lab with entry-level wheeled robots. There are currently three fully integrated tasks:

Drifting is a challenging task in both optimal control and reinforcement learning. It’s a physically unstable maneuver precisely determined by things like vehicle mass, friction, motor torque, mechanics, etc. But, what do you do if you don’t know these things? Simulate and train!

Topics: Dynamics, Controls, Domain Randomization

Vehicle executes a “Scandinavian flick” to drift, which emerged from training!
Vehicle executes a “Scandinavian flick” to drift, which emerged from training!

Elevation intimately connects perception and control. What’s the difference between an obstacle and a ramp? Stairs? For a robot, the answer depends on what kind of thing it is: quadruped, car, drone, etc.

Topics: Spatial Reasoning, Sensor Simulation, Traversibility

Vehicle knows what geometries are walls and what are ramps through trial-and-error in training.
Vehicle knows what geometries are walls and what are ramps through trial-and-error in training.

Visual is a modern simulation-based approach to the age-old problem of navigation with cameras.

Topics: Camera Simulation, Deep Learning & Model Architectures

Cameras are high(huge)-dimensional states that have to be mapped down to actions.
Cameras are high(huge)-dimensional states that have to be mapped down to actions.

Wherever you start, you’ll learn about state-of-the-art simulation and reinforcement learning. And by deploying these models on hardware, you’ll develop valuable engineering skills: hyperparameter tuning, model evaluation & selection, and the killer: solving the Sim2Real gap!