MIT Robotics - Yuval Tassa - Predictive Sampling: Real-time behavior synthesis with MuJoCo

[@pipe_runner_lab]

He made a really good point about GPU vs CPU code and why you might decide to pick one sim over the other.

[@junwencui2370]

I really like Yuval's work and his presentation which inspired me a lot, especially the metaphor of surfing on the cost function.

[@kvasios]

Captivating presentation. MuJoCo MPC is a fantastic project. 56:34 MuJoCo 3 is GPU parallelizable!

[@shauryaseth8859]

Lmoed at the neural network question. Shows how under-appreciated classical control is in a world of "just throw neural net at it"

[@myelinsheathxd]

Amazing! Keep going guys!

[@saliexplore3094]

Wow Yuval is a great presenter. Great job.

[@yokoeric3701]

Exciting! I want to try it now!

[@XecutionStyle]

Incredible. Looking forward to SysID!

[@SiddharthDeore]

Very visual, very fun! ❤

[@varshneydevansh]

Back in 2018 I played a little with the MuJoCo

[@pipe_runner_lab]

Does anyone know which paper he is talking about at 58:49?

[@zippingsugar4075]

Just LOVE IT! I always think RL is overrated in the field of robotics.

[@ZhiboZhang-v2z]

Would you mind to explain this in more detail? I actually need to compare these two method w.r.t motion control

[@justtestingonce]

Come on buddy, stop blaming the Python Gil, there’s many way around it. You can start off with multiprocessing, 1 Gil per cpu.