Hardware Platform

High-performance Computing Units

  • CPU: Intel Xeon / AMD EPYC
  • GPU: NVIDIA A100/H100
  • AI Chips: Google TPU

Edge Computing Devices

  • NVIDIA Jetson series: Orin NX/AGX, 15-275 TOPS
  • Intel NUC
  • Qualcomm RB5

Perception Sensors

  • Vision: RGB cameras, depth cameras, event cameras
  • Environment perception: LiDAR, ultrasonic sensors
  • Proprioception: IMU, force/torque sensors

Actuators

  • Servo motors, stepper motors, direct-drive motors
  • Control interfaces: CAN, EtherCAT, RS485

Simulation Platforms

PyBullet

  • Lightweight, pure Python implementation
  • Physics engine: Bullet

Gazebo/ROS 2

  • Client-server mode
  • Native ROS 2 support

NVIDIA Isaac Gym

GPU-accelerated robot simulation platform, specialized for reinforcement learning training.

Habitat

Facebook developed, indoor visual navigation task simulation.

Open X-Embodiment

1M+ robot trajectories, 22 robot types, 500+ tasks.

Algorithm Frameworks

Deep Learning

  • TensorFlow
  • PyTorch

Reinforcement Learning

  • Stable Baselines3
  • RLlib

Cutting-edge Methods

  • Diffusion Policies
  • Decision Transformer
  • PEFT: LoRA, etc.
  • OpenVLA: Vision-Language-Action model

Optimization Control

  • MuJoCo MPC
  • DiffROS