Real-World ML
In the last couple of years reinforcement learning had several breakthroughs in the field of machine learning. Lots of credit goes to DeepMind who developed alphaGo, alphaZero and more recently alphaStar and alphaFold.
Robots as currently build by Boston Dynamics
Nevertheless in the field of robotics reinforcement learning has not unfold the same amount of acquisition. Robotics hardware is expensive and not accessible to many researchers and engineers around the globe. Furthermore it is harder to train real-world reinforcement learning robots since you can not simulate multiple worlds in parallel. Marc Raibert the CEO of Boston Dynamics lately claimed that they do not use any reinforcement learning in their robots. It is suspected that Boston Dynamics mostly relies on sequential composition of dynamically dexterous robot behaviors. The vision of this project is to push the field of real-world reinforcement learning more into the direction of its boundaries. In the future complicated walking behaviors and motion control may uses more reinforcement learning as these fields are using it right now.
To accelerate the process of reinforcement learning in the field of robotics realworldML developed an open-source six legged robot which can be 3D-printed from anyone around the globe. The costs to build a fully working hexapod does not exceed a 200 Euro threshold.