China’s A.I. Powered Robot Nearly Fights Back After Being Hit, Pulled By Trainer 

This is not investment advice. The author has no position in any of the stocks mentioned. has a disclosure and ethics policy.

Chinese robotics technology firm LimX Dynamics’ Biped Robot P1’s latest demonstration test shows the machine’s ability to respond remarkably well when threatened by a human and traverse complex terrain. The robot uses reinforcement learning to respond to outside stimuli such as moving objects or bumps on a path, according to the firm. Reinforcement learning is a subset of a set of techniques that are used to train artificial intelligence algorithms, and the latest news about the firm, also called Zhuji Dynamics, builds up on its steady development of robotic technologies.

LimX Dynamics’ Biped P1 Robot Successfully Navigates Mountainous Forest Terrain Via Reinforcement Learning

LimX is a robotics company with a humanoid robot, a quadrupled robot and the biped P1 in its product portfolio. The latest test run for the P1 platform saw the firm place the P1 in the Tanglang Mountain in Shanghai, China. The purpose of the test was to test the platform in an unpredictable environment characterized by rough terrain that required the P1 to be continually aware of its surroundings and incorporate the data back into its responses through reinforcement learning.

LimX shared details of the test in a Medium post. According to the firm, one of the most complex aspects of the test was that, unlike urban environments, the mountains do not have uniform steps or ground level. The P1’s test marked a third test for the firm’s full product portfolio, with earlier tests involving humanoid and quadrupled robots. The latter took place in October last year, and it saw the quadrupled X1 slowly make its way around a steel factory in South China.

As it did for its previous tests, the firm also shared a video of the P1’s trekking trip. The test showed the robot climbing grass hills, moving around ditches, recovering its gait after tripping and not only remaining stable after being kicked and pulled but also maintaining stability after repeated hits to the legs with a stick. Close inspection of the footage also revealed the trainer’s stick breaking after it hit the robot’s legs.

The LimX Dynamics P1 rapidly covers rocky terrain in a test. Image: Screengrab from LimX Dynamics’ Biped Robot P1 Conquers the Wild Based on Reinforcement Learning on YouTube

While LimX Dynamics did not share any details of the distance that the P1 covered during its maiden mountainous run, it did provide some details about the robot’s training through reinforcement learning.

Unlike other machine learning branches, such as supervised and unsupervised learning, which rely on or eliminate data tags to train software, reinforcement learning uses rewards or penalties to enable a system to make decisions. Machine learning systems differ from artificial intelligence systems in their reliance on autonomous learning, but they are categorized under the broader umbrella of A.I.

LimX outlined that while the test did not use any data related to its environment to “feed” the P1, the robot nevertheless managed to navigate its complex test environment. The firm hopes to apply similar technologies to its humanoid robot, the CL-1. The humanoid completed a test in late December, which saw it climb stairs and move indoors and outdoors. This test saw the CL-1 incorporate data from stairs and other objects before successfully navigating through them.

You can take a look at the P1’s mountain test in the video below.

Share this story