1. Researchers at MIT have developed a two-part control system that improves the speed and agility of legged robots as they jump across gaps in terrain, without requiring the terrain to be mapped in advance. The system uses a camera mounted on the front of the robot to capture depth images of the upcoming terrain, which are fed to a high-level controller along with information about the state of the robot's body.
2. The researchers used reinforcement learning to train the high-level controller, conducting simulations of the robot running across hundreds of different discontinuous terrains and rewarding it for successful crossings. They then tested their control scheme using the MIT mini cheetah, which successfully crossed 90% of terrains.
3. While the researchers were able to demonstrate that their control scheme works in a laboratory, they still have a long way to go before they can deploy it in the real world. They hope to mount a more powerful computer to the robot so it can do all its computation on board, improve its state estimator to eliminate the need for motion capture systems, and enhance both low-level and high-level controllers.