AI agents are learning how to do all kinds of interesting jobs, including the creative jobs we like to handle ourselves. Nevertheless, technology advances.YouTuber is active in this field [AI Warehouse]has taught them to walk in environments simulated by artificial intelligence.
The AI controls a vaguely humanoid-like creature, albeit with a greatly simplified body and limbs. It “lives” in a 3D environment created with the Unity engine, which provides the physics engine you need to work with. Meanwhile, the ML-Agents package is used to provide his Albert’s brain, the AI responsible for learning to walk.
In this video, we will step through various “Deep Reinforcement Learning” tasks. These reward you for completing goals designed to teach the AI how to walk. Albert is given limb control and tasked with just reaching a button on the floor a short distance away. After many trials, he learned the worm’s ways and achieved his goal.
Getting Albert to walk upright required a lot more training. The uneven ground and wall between him and the goal was exploited to overcome the challenge, and he was encouraged to alternate between his legs and maintain an upright posture. Over time, he was able to progress through skipping to a near-perfect gait cycle.
Some might argue that this method of teaching requires a lot of specific instruction, but it is still an admirable feat to accomplish. That’s arguably more outright complicated than learning to play Trackmania, but it was impressive enough on its own. Video after break.
