Learning Machines from Simulation to Real World
This addresses the simulation-to-reality gap for robotics applications, though it appears incremental as it applies existing methods to new tasks.
The paper tackled the problem of transferring learned behaviors from simulation to real-world robots, demonstrating that a reinforcement learning agent trained in a virtual environment can reliably perform tasks like obstacle avoidance, foraging, and predator-prey scenarios in unknown real environments.
Learning Machines is developing a flexible, cross-industry, advanced analytics platform, targeted during stealth-stage at a limited number of specific vertical applications. In this paper, we aim to integrate a general machine system to learn a variant of tasks from simulation to real world. In such a machine system, it involves real-time robot vision, sensor fusion, and learning algorithms (reinforcement learning). To this end, we demonstrate the general machine system on three fundamental tasks including obstacle avoidance, foraging, and predator-prey robot. The proposed solutions are implemented on Robobo robots with mobile device (smartphone with camera) as interface and built-in infrared (IR) sensors. The agent is trained in a virtual environment. In order to assess its performance, the learned agent is tested in the virtual environment and reproduce the same results in a real environment. The results show that the reinforcement learning algorithm can be reliably used for a variety of tasks in unknown environments.