Explainable Robotics applied to bipedal gait development
Nico Roos, Zhenglong Sun · Research Publications (Maastricht University) · 2019
Explainability is becoming an important topic in artificial intelligence (AI). A well explainable system can increase the trust in the application of that system. The same holds for robotics where the walking gait controller can be some AI system. We will show that a simple and explainable controller that enables an energy efficient walking gait and can handle uneven terrains, can be developed by a well structured design method. The main part of the controller consist of three simple neural networks with 4, 6 and 8 neurons. So, although creating a stable and energy efficient walking gait is a complex problem, it can be generated without some deep neural network or some complex mathematical model.