The robot from Boston Dynamics surprise with their ability to move independently and a recent video shows them dancing, a real demonstration of agility. However, these robots took years of development to get there. Developing robust walking controllers for bipedal robots is a challenging endeavor.
A group of researchers from the University of Berkeley (California) has opted for a new, faster method that allows a two-legged robot to learn to walk on its own. A pair of robot legs called Cassie, a design by Agility Robotics, has been taught to walk using reinforcement learning, the training technique that teaches AIs complex behavior via trial and error.
Many of the videos that you see of virtual agents are not at all realistic. These videos can make people believe that this is a simple problem and that it is already solved. But we still have a long way to go to teach humanoid robots to operate safely in human environments.
The heavy two-legged robot can lose balance and fall if its movements are even a tiny bit off. Therefore, training a large robot through trial and error in the real world would be dangerous. To get around these problems, the Berkeley team used two levels of a virtual environment.
In the first, a database of steps from different sources was loaded into its system to simulate different movements. This simulation was then transferred to a second virtual environment called SimMechanics that mimics real-world physics with a high degree of accuracy.
After testing in those virtual environments, the information was transferred to the robot and tested in a real environment without any extra fine-tuning. Cassie managed to perform a set of diverse and dynamic behaviors while also being more robust than traditional controllers and prior learning-based methods that use residual control. It performed versatile walking across rough and slippery terrain, carried unexpected loads, and recovered from being pushed. The bipedal robot was even able to continue walking after two motors failed in its right leg.
Cassie is not yet ready to compete with Boston Dynamics’ Atlas or Spot, but this advance should help accelerate the development of new robots.
