Marius Memmel

@memmelma

Robotics PhD student @UW and intern @amazon, previously intern @NVIDIA, intern @Bosch_AI, @EPFL, @TUDarmstadt, @DHBW

Seattle, WA

memmelma.github.io

Replies Reposts Media

Pinned
Marius Memmel
@memmelma
Mar 5
There’s a discussion going on rn about two recent robotic reward models: TOPReward⛰️ and Robometer🌡️ Which one is better? It depends entirely on your objective! Here is a deep dive into the conceptual differences, strengths, and weaknesses of both. 🧵👇
3
Marius Memmel
@memmelma
Sep 3
Update the high-level VLM while keeping the low-level policy fixed. Then run RL on offline data to avoid expensive environment interactions. Clever idea!
Lehong Wu
@LehongWu2004
Aug 28
How should robots think? 🤔 In our new paper R³, we introduce a simple way to get robots to think before they act—like humans: free-form, deliberate, and actually useful. The recipe is surprisingly simple: just RL, done carefully. Yet it enables reasoning that can effectively
00:00
Marius Memmel
@memmelma
Aug 27
I recently tried out their github repo on some random mujoco environment and was blown away by how well a zero-shot VLM could already solve the task. No policy, no fancy action head, just outputting joint positions or endeffector actions! Highly recommend checking it out :)
Jay Chooi
@chooi_jeq
Aug 27
Introducing @robocurve, a Public Benefit Corporation to measure and report frontier robotics capabilities. We build real-world evaluations for robots and publish results as a neutral third party.
00:00
2
Marius Memmel
@memmelma
Aug 21
When we were exploring trajectory retrieval, we were hoping that some day in-context learning would be unlocked for robotics. Turns out all it took was two years and tons and tons of data. Congrats @GeneralistAI on the impressive demo! Check out our past work on retrieval:
Generalist
@GeneralistAI
Aug 19
Introducing GEN-1.5, a one-shot learner. It can learn new tasks in a few seconds. Show it what to do, and it generalizes. This capability emerged from pretraining on physical data at scale, as a step towards our mission of building general intelligence for the physical world.
00:00
1
Marius Memmel
@memmelma
Jul 2
Damn, this policy SCOREs more goals than the German team at the World Cup ⚽️
Abhishek Gupta
@abhishekunique7
Jul 2
Replying to @abhishekunique7
SCORE gets the benefits of simulation for policy improvement: parallel interaction, privileged state, resets, and robustness to perturbations. Meanwhile, it avoids much of the manual engineering usually needed to make sim RL work, such as dense reward shaping, curriculum
00:00
2