Loading activity…
@djiafei
Assistant Professor at @NUScomputing | Robotics & AI PhD @uwcse | Host of @RoboPapers | Ex-@allen_ai robotics lead, @NVIDIA | Magician | my opinion is my alone.
AI
A Reflex developer says the robot has about one second to perceive, predict and move its whole body to act.
AI
A user describes a video comparing GPT-6 Astra, Opus 5.5, Grok 4.7 and MolmoAct2 as unsped and not cherry-picked. They say LLM robot control is progressing but has a way to go.
AI
The announcement proposes building robot foundation models on a pretrained 3D grounding model, while questioning whether language and video models are the right starting point.
AI
A hiring announcement seeks candidates with experience in robotics hardware, robot learning or post-training.
AI
The post argues that learning movement without images can avoid visual shortcuts. Later vision training is meant to support the action expert’s existing spatial understanding.
AI
The user says they borrowed code to run an open-loop rollout on a trajectory from the MolmoAct2 dataset.
AI
The post points to harness-design research as a way to improve performance, highlighting Graph-as-Policy (GaP), which it says was accepted to CoRL.
AI
@scaling01, operator of LisanBench benchmark, made the statement about future AI models.
AI
Assistant professor at NUS Computing posts about recent robotics and AI paper acceptances.
AI
Podcast hosts share retrospective site with stats from embodied AI discussions.
AI
The foundation model performs unseen long-horizon tasks through in-context learning.
