Google researchers have developed an Artificial Intelligence (AI) system that imitates the motion of animals and trains robots for better agility. The researchers trained an 18 DoF quadruped robot named Laikago to perform a variety of ‘tricks’ aka agile behaviours such as dynamic hops and turns, and gaits. The researchers released a preprint paper and an elaborate blog post announcing this advancement early this week. 

While previously, legged-robots did find success in learning motion, fluid and agile locomotion skills and stability on various surfaces were always a challenge. The researchers captured about 200 million sample data sets of a dog’s motion via sensors and deployed reinforcement learning technique (RL) to ‘train’ Laikago. The researchers used the datasets to create a stimulated dog animation and matched key points of its motion and that of Laikago, the robot, to replicate the exact same movement as the animal. The RL technique rewards robots for successfully completing tasks. The result was that the AI system has ‘trained’ Laikago more successfully than traditional hand-coded robotic controls. “We show that by leveraging reference motion data, a single learning-based approach is able to automatically synthesize controllers for a diverse repertoire [of] behaviours for legged robots,” wrote the coauthors in the paper. “By incorporating sample efficient domain adaptation techniques into the training process, our system is able to learn adaptive policies in simulation that can then be quickly adapted for real-world deployment.”

Finally, the researchers were able to port the final control algorithm into Laikago in the lab. Laikago learned to stabilize movements and correct for differences in weight distribution and design — though some moves weren’t entirely successful. “Due to hardware and algorithmic limitations, we have not been able to learn more dynamic behaviours such as large jumps and runs. Exploring techniques that are able to reproduce these behaviours in the real world could significantly increase the agility of legged robots. The behaviours learned by our policies are currently not as stable as the best manually-designed controllers,” wrote the researchers in the blog. But as technology progresses, this innovation can pave the path for the creation of goods transport robots that can help with transporting materials between multilevel warehouses and fulfilment centres across the globe.