Utilizing AMD Schola and UnrealRoboticsLab with AMD ROCm™ Software to Train a Robotic Arm

2026-06-17 · Source: AMD ROCm Blogs · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, extended

Summary

This post details a pipeline for training a 6 degrees-of-freedom (DOF) robotic arm using AMD Schola, UnrealRoboticsLab (URLab), and AMD ROCm™ software within Unreal Engine 5.7. The system integrates Unreal Engine's high-fidelity rendering with MuJoCo's accurate contact physics via URLab, which imports robots defined in MJCF. AMD Schola, an open-source Unreal plugin version 2.1.1, then exposes this environment to Python training frameworks like Stable Baselines 3 through gRPC, creating a standard Gymnasium-compatible interface. The article provides bundled C++ code for an xArm6 reach task, covering observation and action space definitions, reward shaping, episode termination logic, and observation normalization. Training can be accelerated on AMD GPUs using PyTorch on ROCm, supporting multiple parallel simulators for increased throughput. The pipeline allows for iterative development, live policy visualization, and deployment of ONNX models.

Key takeaway

For AI Engineers developing robotic arm control policies, this pipeline offers a robust solution for high-fidelity simulation and training. You should utilize Unreal Engine 5.7 with URLab for accurate physics and Schola for Gymnasium-compatible RL loops. Configure reward shaping and observation normalization within `AXArmEnvironment` and use AMD ROCm™ for GPU-accelerated training. This approach streamlines development, allowing visual debugging and efficient deployment of ONNX models.

Key insights

Integrating Unreal Engine, MuJoCo via URLab, and Schola enables high-fidelity, physics-accurate robotic arm reinforcement learning on AMD hardware.

Principles

• Combine high-fidelity rendering with accurate physics.
• Decouple simulation from RL training via gRPC.
• Pin observation/action spaces in code for consistency.

Method

Define robot in MJCF, import into Unreal via URLab, integrate C++ glue code for Schola, configure reward/termination/normalization, then train using Schola CLI with Python frameworks.

In practice

• Use `AXArmEnvironment` for custom task logic.
• Disable onboard camera streaming for faster training.
• Clone environment actors for parallel training.

Topics

• Robotic Arm Training
• Reinforcement Learning
• Unreal Engine Simulation
• MuJoCo Physics Engine
• AMD ROCm Software
• Schola RL Plugin

Code references

• GPUOpen-LibrariesAndSDKs/Schola
• URLab-Sim/UnrealRoboticsLab
• xArm-Developer/xarm_ros
• ROCm/TheRock

Best for: Robotics Engineer, Machine Learning Engineer, AI Engineer

Related on AIssential

• Training a Robotic Arm Using MuJoCo and JAX on AMD Hardware with ROCm™ — Reinforcement learning with MuJoCo and JAX on AMD ROCm enables robust robotic arm training for complex manipulation tasks.
• Edge-to-Cloud Robotics with AMD ROCm: From Data Collection to Real-Time Inference — AMD's edge-to-cloud robotics solution integrates LeRobot, VLA models, and Imitation Learning on AMD hardware for efficient robot training and deployment.
• NVIDIA launches Newton 1.0 physics engine for industrial robot training — NVIDIA has released Newton 1.0, a production-ready open-source physics simulation engine designed for training robots in contact-rich manipulation and locomotion tasks, built on NVIDIA Warp and OpenUS…
• NVIDIA’s New AI Shouldn’t Work…But It Does — DreamDojo enables robots to learn complex real-world physics and interactions from human video data.
• [P] I open-sourced a synth framework for creating physics-simulated humanoids in Unity with MuJoCo -- train them with on-device RL and interact in VR — This system enables physics-based humanoid creation and on-device reinforcement learning for mixed reality.
• Scale Robot Reinforcement Learning with NVIDIA Isaac Lab on Amazon SageMaker AI — Scale robot RL training efficiently by offloading compute infrastructure management to Amazon SageMaker AI.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AMD ROCm Blogs.