Mohak Bhardwaj

Research Scientist at Boston Dynamics | Ph.D. CSE @uwcse | M.S Robotics Systems Development from @CMU_Robotics

[Jayjun Lee @jayjunleee]

What if robot hands could acquire dexterity once, then reuse it to learn downstream tasks much faster?

Introducing ADEPT: a pre-training & post-training paradigm using RL entirely in sim, then zero-shot deploying visuo-tactile policies in the real world.

🔗https://t.co/jdIR1CnG4c with Dex team @NVIDIAAI

[Rohan Baijal @RohanBaijal]

Some more 'cherry' picked field test adventures with Spot! :)

[The Humanoid Hub @TheHumanoidHub]

Boston Dynamics collaborated with NVIDIA to demonstrate DextrAH-RGB, a workflow for dexterous grasping from stereo RGB input.

The end-to-end policy for Atlas robot, trained entirely in NVIDIA Isaac Lab, transfers zero-shot from simulation to the real robot.

[Nathan Ratliff @robot_trainer]

Our work with Boston Dynamics on RL training workflows for Atlas (upper torso) manipulation. An amazing team to work with!

Gina Fey, @mostlymohak @_mlutter, Alberto Rodriguez
@ritvik_singh9, Karl Van Wyk, @ankurhandos

[Boston Dynamics @BostonDynamics]

Wishing you a holiday season full of light and laughter as we flip over into the new year!

[Boston Dynamics @BostonDynamics]

Atlas is autonomously moving engine covers between supplier containers and a mobile sequencing dolly, using ML to detect and localize the environment fixtures and individual bin. There are no prescribed or teleoperated movements. https://t.co/br63dMoB7C

[Boston Dynamics @BostonDynamics]

Atlas doing a quick warm up before work.

[Mohak Bhardwaj @mostlymohak]

@hganjoo_153 Congrats bhai!!

[Mohak Bhardwaj @mostlymohak]

Come over to the poster session 4 at @l4dc_conf today if you're interested in learning more about our work on controlling a complex fluid dynamical system with RL directly in the real world.
https://t.co/SkVj5pPoHN

[Markus Wulfmeier @m_wulfmeier]

.@l4dc_conf is happening @UniofOxford this week!
Two fantastic papers coming up from our group @GoogleDeepMind in collaboration with @MIT!

First @timseyde continues the path of 'Q-learning is all you need' even for continuous control with 'Growing Q-Networks'
https://t.co/QDnFa5BPOe

Second @mostlymohak demonstrates RL for complex, real-world applications on the 'Box o’ Flows'
https://t.co/bnDhuagCUY

[Mohak Bhardwaj @mostlymohak]

@m_wulfmeier @BostonDynamics Thanks Markus!

[Mohak Bhardwaj @mostlymohak]

Slightly belated life update: In April I started as a Research Scientist in the Atlas team at @BostonDynamics. Really excited to continue research in learning and manipulation with this amazing team and our shiny new humanoid!

[Mohak Bhardwaj @mostlymohak]

@harshit_sikchi @BostonDynamics Thanks! @harshit_sikchi

[Mohak Bhardwaj @mostlymohak]

@rwik_rana @BostonDynamics Thanks Rwik!

[Mohak Bhardwaj @mostlymohak]

@hganjoo_153 @BostonDynamics Thanks bhai!!

[Mohak Bhardwaj @mostlymohak]

@psinghal8 @uwcse Thanks! Yeah it does seem like ages have passed

[Mohak Bhardwaj @mostlymohak]

Really excited to share that I successfully defended my PhD thesis at @uwcse earlier this month! My deepest gratitude to my committee and everyone who helped me along this 5 year journey!

[Sanjiban Choudhury @sanjibac]

Cooking in kitchens is fun. BUT doing it collaboratively with two robots is even more satisfying!

We introduce MOSAIC, a modular framework that coordinates multiple robots to closely collaborate and cook with humans via natural language interaction and a repository of skills.

[Mohak Bhardwaj @mostlymohak]

A big thanks to my collaborators Jonas Buchli, Martin Riedmiller, @m_wulfmeier, Thomas Lampe, Michael Neunert, Francesco Romano, Abbas Abdolmaleki, Arunkumar Byravan and everyone else on the Controls team for their support.

[Mohak Bhardwaj @mostlymohak]

Really excited to share work done during my @GoogleDeepMind internship. We introduce Box o' Flows, a benchtop control system for real-world fluid directed rigid body control and demonstrate how deep RL can be used to learn highly dynamic tasks directly on real hardware.

[Markus Wulfmeier @m_wulfmeier]

Reinforcement learning is most useful if a) demonstrations are hard to get and b) a system is hard to model.

( a) no imitation, b) no MPC etc)

Excited to share Mohak's internship report and the 'Box o Flows' enabling us to ask questions in this space!

https://t.co/SAHBr2OBal

[Mohak Bhardwaj @mostlymohak]

Full version of our @NeurIPSConf 2023 paper titled Adversarial Model for Offline Reinforcement Learning is now available on arxiv https://t.co/qxJMqQsDfJ
Joint work with @tengyangx , @nanjiang_cs, Byron Boots and @chinganc_rl. Open source code coming soon!