Goals: 3 Years to develop NCKU+ASR Humanoid robot Small Brain (auto walking, balancing, …)

1st year: •Nikolas: use any Humanoid URDF and Reinforcement Learning methods to make the humanoid walk, fall, and learn. Use the framework with most feasibility, most resources to replicate, learn, and develop. Reference: NVIDIA, Boston Dynamics, Unitree, MIT, Agility, other possible open-sources, produce a standard working solution >> transfer to MingYang •MingYang: use Nikolas solution, for ASR Humanoid model

2nd year: •NIKOLAS: use NCKU robot URDF •MingYang: use Nikolas solution, for both NCKU and ASR Humanoid URDF

3rd year: sim to real, NCKU Humanoid integration, ASR Humanoid integration (BODY+HAND+FINGERS+LEGs)

Reviews of State-of-the-art Humanoid Robot Platforms in 2024 Our Project Goal Tasks Theoretical: Robotics, Control, Design, Mechanical, Electrical, Reinforcement Learnging, … Design: Inventor, Solidworks, OnShape, Manufacturing: Simulations: NVIDIA Omniverse, Issac Gym, Issac Lab, and Issac Sim Physical: Mechanical, Electrical, Control, CS, Software: ROS, Issac ROS, Computing platform: AGX, CUDA, RTX, DGX, … ASR + NCKU + Delta Cooperation Required Resources Expected Timeline Q&A

PRE

Introduction to Issac Gym and Issac Sim

Gym is for RL/ML Engineer with understanding in MLOPS

Sim is for Simulation Engineer with understanding of the use cases and actual worlds physics of the robot environments

Reinforcement Learning Framework and how Gym and Sim help to accelerate it

Application for Nvidia Project GR00T

Requirement

Resources

Human

PLAN

Year 1

Nikolas:

  1. Understand Theory of Reinforcement Learning in Simulation Environments

    1. https://short-horizon-actor-critic.github.io/

    https://www.youtube.com/watch?v=wqvAconYgK0&t=166s

  2. Test publicly available URDF/UDF/USD Model

    1. https://docs.omniverse.nvidia.com/isaacsim/latest/features/environment_setup/assets/usd_assets_robots.html