Michael Cho ✈️ CoRL

This account is verified. Learn more

@micoolcho

I ❤️ robots, cheap hardware, steam engines, XGBoost, Liverpool FC & SG 🇸🇬 | Plane crash survivor | Building @BitRobotNetwork @frodobots

Singapore Joined May 2010

2.7K Following 6.1K Followers 3.1K Posts

Latest posts 5

Pinned post

Michael Cho ✈️ CoRL @micoolcho 4d

Real-world in-the-wild, multiple embodiments evals at scale!

Really hope that @RoboColosseum can move the needle for the industry!

Huge team effort from @huggingface @eaic_sg @BitRobotNetwork folks...special shoutout to @DJiafei @ruijie_sg @pepijn2233 ; we have a ton to do!

RoboColosseum Verified account @RoboColosseum 4d

Introducing RoboColosseum

An open robot eval initiative by @huggingface @BitRobotNetwork @eaic_sg @DJiafei

Access:
- 100+ robots (@UnitreeRobotics G1s, @LeRobotHF SO-101s, YAMs, Franka and more)
- 140+ in-the-wild scenes
- Open rollout data
- 10,000 FREE eval hrs

Thread ↓

View quoted post on X

Michael Cho ✈️ CoRL @micoolcho 1h

Deciding what to put on the X/Y axis is 99.9% of the art of plotting great graphs. @sebkrier nailed it here.

Séb Krier Verified account @sebkrier 4h

Updated version!

View quoted post on X

Michael Cho ✈️ CoRL @micoolcho 4h

RoboCap + Matrix Gloves = Tactile Touch

Come take a look at our booth at CoRL! Super excited for this collab with our friends at Matrix!

BitRobot 🦾 Verified account @BitRobotNetwork 10h

1/ RoboCap data is getting a sense of touch 👋

We’re adding tactile sensing with @moxianmxkj MX Matrix Gloves compatibility.

Together, they provide hand contact and pressure data to visual RoboCap demos, giving researchers more context for dexterous robot learning. ↓

View quoted post on X

Michael Cho ✈️ CoRL @micoolcho 10h

There r lots of positive signals from negative outcomes (ie. learning from mistakes).

Congrats @fancy_yzc & team on the cool work!

Zhecheng Yuan Verified account @fancy_yzc 11h

4/8
Learning from negative experience makes a measurable difference.
Excluding negative-advantage samples reduces zip-tie success from 86% to 60%. Removing the trust-region constraint causes training to diverge.
With the same collected experience, TRFO also outperforms HG-DAgger and RECAP on laundry folding and zip-tie threading. Both the negative learning signal and the constraint on its strength contribute to effective policy improvement.

View quoted post on X

Michael Cho ✈️ CoRL @micoolcho 15h

Starting to see these real jumps in performance! Agentic robotics ftw!

Robotics Center of Silicon Valley @RoboticsCtr 2d

0% success on L3 tasks in a recent Astra benchmark, including bottle-cap unscrewing.

Now RAX tackles jar opening with real dexterous hands.
6 chained skills. 4 twists. 140° rotation.

The jar opens.

@RunhanH @xczhangxx @du_yilun
#RAX #Robotics

View quoted post on X