Simulated robots can now spin-kick like a karate expert or backflip like an acrobat. The Berkeley Artificial Intelligence Research (BAIR) Lab yesterday proposed DeepMimic, a Reinforcement Learning (RL) technique that enables simulated characters to regenerate highly dynamic physical movements learned from data collected from human subjects. BAIR is a top-tier research lab focused on computer vision, machine learning, natural language processing, and robotics.
RL methods have been shown to be applicable to a diverse suite of robotic tasks, particularly motion control problems. A typical RL includes a policy function that consists of all action selections that machines can do, and a value function that returns a low or high reward each time a machine takes an action. Machines can self-learn skills by leveraging the reward. The epoch-making Go computer AlphaGo produced by DeepMind is grounded on the same technique.
However, virtual characters trained with deep RL can exhibit abnormal behaviours such as jittering, asymmetric gaits, or excessive movement of limbs.
BAIR’s new paper DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skill introduces a policy function that collects challenging skills such as locomotion, acrobatics, martial arts, and dancing.
BAIR next initialises a character to a state sampled randomly, a method known as Reference State Initialization (RSI). The character can learn skills from any state of moves, such as the inflection point of a flip, and RSI can allow the character to know which states will result in high rewards even before it has acquired the proficiency to reach those states.
By connecting RSI with Early Termination (ET), a standard practice for RL researchers to stop simulations that lead to failure, BAIR researchers ensured that a substantial proportion of the dataset consists of samples close to the reference trajectory. Without ET, the character may flail or fall, but will not learn to flip.
The research shows that the character can learn over 24 skills, with movements nearly indistinguishable from the human reference subjects. BAIR also says its technique is simpler and produces better results than the current leading motion imitation method, Generative Adversarial Imitation Learning (GAIL).
BAIR hopes the new research will facilitate the development of more dynamic motor skills for both simulated characters and robots in the real world.
Author: Paul Fan| Editor: Tony Peng, Michael Sarazen
Pingback: BAIR Open-Sources Popular DeepMimic Project | Synced
Pingback: BAIR Open-Sources Popular DeepMimic Project – AI+ NEWS
I suppost you to raise this idea.
Unlike clunky, jittery virtual models of the past, DeepMimic’s characters move fluidly, Ragdoll Hit almost indistinguishably from real athletes, thanks to a clever blend of physics-based simulation and example-guided learning.
I find it very interesting
By connecting RSI with Early Termination (ET), a standard practice for RL researchers to stop simulations that lead to failure, BAIR researchers ensured that a substantial proportion of the dataset consists of samples close to the reference trajectory.
This technology can change the way we interact with computers and the environment around us, opening the door to new and innovative experiences. baseball 9
The character can learn skills from any state of moves, such as the inflection point of a flip, and RSI can allow the character to know which states will result in high rewards even before it has acquired the proficiency to reach those states.
The sports games research indicates that characters trained using this combined approach can learn over 24 distinct skills. The movements exhibited by these characters are reported to be nearly indistinguishable from those of human reference subjects, showcasing the effectiveness of the technique.
Happy holidays my beloved friends! I have the pleasure of presenting to you a game that is a Slope Run, which is a very intriguing game. This is a really appealing game that involves running indefinitely. While you are engaged in difficult and exhausting activities like studying and working, it assists you in successfully relieving tension.
Impressive work! The combination of RSI and ET seems particularly clever for efficient skill learning.
This game quordle has become a part of my daily routine, and I look forward to it every day. It’s a perfect way to unwind while keeping my mind sharp. What do you all do to incorporate word games into your routine?
It’s going to get cold, so get ready for Geometry Dash SubZero, a brand-new Geometry game with three tough and different levels!
Freaky Font is a free online tool that transforms normal text into stylish, creepy, cute, and fun Unicode fonts you can copy and paste anywhere.
Berkeley researchers create a virtual acrobat by developing advanced simulation and machine learning techniques that allow digital characters to perform complex, Delta Executor physics-based movements with agility and balance, opening new possibilities in animation, robotics, and understanding human motion.
Freaky Font is a free online tool that transforms normal text into stylish, creepy, cute, and fun Unicode fonts you can copy and paste anywhere.
The way DeepMimic handles reference motion is insane—much more fluid than traditional RL approaches. It’s like finding an escape road 2 the ‘uncanny valley’ of character physics. I’m curious if this framework can be adapted for real-time robotic gait correction next. Solid breakdown of the Berkeley paper!