AMD: Open-source Schola connects Unreal Engine and reinforcement learning for robotic arm training on ROCm

AMD published a tutorial for Schola, an open-source plugin that connects Unreal Engine with reinforcement learning for training robots on AMD hardware.

A bridge between Unreal and RL frameworks

Schola is an open-source plugin for Unreal Engine that enables reinforcement learning (RL) training compatible with Gymnasium, through Python frameworks connected via gRPC. Reinforcement learning is a method in which an agent learns by trial and error, maximising a reward signal. By connecting Unreal with tools like Gymnasium, Schola bridges realistic simulation with established RL libraries.

Training a robotic arm on the AMD stack

In the example, a collaborative robotic arm xArm6 (with six degrees of freedom) is trained in Unreal Engine 5.7 with MuJoCo physics. The algorithm is PPO (Proximal Policy Optimization) via Stable Baselines 3, with computation handled by PyTorch on the AMD ROCm stack for GPU acceleration. The task is a “reach” — the arm tip moves to randomly placed target locations. The tutorial is a practical guide without quantitative benchmark results, serving as a toolchain demonstration rather than a performance comparison.