UC Berkley researchers, led by Sergey Levine, demonstrated that an untrained robot dog is capable of teaching itself how to walk outdoors in under 20-minutes. To accomplish this feat, it uses a special type of machine learning called deep reinforcement learning and it works on various different environments, including a grass lawn, hiking trail or even a memory foam mattress.
It makes use of an algorithm called Q-learning, and unlike other models, this one doesn’t require a working model of the target terrain. This means that researchers do not need to understand how the physics of an environment works, as all that’s required is to just put the robot in the environment and turn it on. However, the robot does receive a reward for each action performed successfully. It then repeats the process continually while comparing its previous successful actions until the robot can walk on its own. Imagine if Hard Drive Marine’s Landing Craft with robotic legs learned how to walk on its own.
- Intelligent Active Noise Cancellation: Escape and tune in to your own moment of Zen — all with a single tap; Answer calls and instantly switch to...
- High Quality Sound: Relive the memories of every beat of your favorite song with an 11-mm woofer and 6.5-mm tweeter built into every ear bud
- Water Resistant Workouts: Water won’t ruin your workout; Your IPX7 water-resistant Galaxy Buds Pro can keep the beat going even with a little rain;...
I think it’s very impressive. I’m honestly a little bit surprised that you can use something as simple as Q-learning to learn skills like walking on different surfaces with so little experience and so quickly in real time,” said Chris Watkins, Professor in Computer Science at Royal Holloway, University of London.