EdgeRL: A Light-Weight C/C++ Framework for On-Device Reinforcement Learning

초록

Advances in reinforcement learning (RL) have achieved significant success in many areas. However, RL typically requires a large amount of computation and memory. Often RL implemented in Python is too heavy to run on a resource-limited edge device. Therefore, making the RL model lighter is very important for on-device machine learning. In this paper, we propose a lightweight C/C++ RL framework aiming for RL on edge devices. The proposed RL framework is designed to run on a single-core processor that is typically included in a resource-limited embedded platform. The evaluation using OpenAI Gym's CartPole demonstration shows that the model can be trained on an edge device in real-Time.

키워드