The fastest robot.
Using advanced reinforcement learning techniques, the robot can alternate between walking driving modes adjusting to the different terrain it encounters. The main objective of the project was to build a large-scale autonomous driving system for this terrestrial robot with the fastest speed of all time, said Joonho Lee co-author of the paper published in the project.
Building on the advances of the previous robots created by the Serveras team that won the Darpa challenge in 2021, the new robot is capable of performing complex tasks with high efficiency, it can move at up to 20 km per hour an impressive speed that required the creation of Fast and accurate navigation plans to avoid collisions.
Testing.
To allow the robot to navigate autonomously, the researchers developed hierarchical reinforcement learning methods, training a controller based on a neural network capable of handling various inputs and generating navigation plans in milliseconds.
In tests carried out in real environments, the navigation system proved to be highly responsive and efficient, allowing the robots to travel more than 10 km in two different European cities Zurik and Seville. Robots with wheels are efficient, but robots with wheels cannot cross high obstacles. Legs are good at overcoming obstacles, but have low efficiency, Lee said.
With the combination of legs and wheels the robot can overcome obstacles more efficiently and operate for longer, normally walking robots operate for up to an hour, but the ANYmal can operate at least three times longer.
