Angjoo Kanazawa

@akanazawa

Associate Professor at @Berkeley_EECS, @berkeley_ai. KAIR, @nerfstudioteam. Amazon Scholar @ FAR. Prev advised @WonderDynamics and @LumaLabsAI. she/her.

following
followers

EyeRobot 2.0 uses active gaze instead of wrist cameras for robot manipulation

A post about EyeRobot 2.0 says its gaze enables precise manipulation without wrist cameras, which tools can block.

1 Source, 1d ago, first seen 1d ago

QF3 introduces an off-policy update for flow-policy reinforcement learning

A researcher says QF3 can train humanoids from scratch and fine-tune VLAs and image models with one update.

1 Source, 6d ago, first seen 6d ago

PRISM turns four real videos into 256 counterfactual variants for humanoid robot training

A PRISM team member says the project recovers robot-object trajectories and trains one policy that generalizes across objects, scales and layouts.

1 Source, 15d ago, first seen 15d ago

ABC robotics stack released with a 130K-plus-episode teleoperation dataset

A post announcing ABC’s release says its dataset spans 3,500 hours and 195 tasks, collected on an $8,000 bimanual setup. It says the hardware, training code, simulation and evaluation tools are open too.

1 Source, 20d ago, first seen 20d ago

The prospect of large-scale real-to-simulation systems

A user sees large-scale real2sim—bringing the real world into simulation—as within reach, noting how hard it once was to imagine OpenAI putting serious resources into 3D/4D.

11 Sources, 34d ago, first seen 34d ago

The case for language models at the core of physical AI

A researcher describes work across Gemini, Omni and generative media to test ideas about physical AI, arguing that language models remain indispensable.

16 Sources, 29d ago, updated 29d ago

Featured in

EyeRobot 2.0 uses active gaze instead of wrist cameras for robot manipulation

QF3 introduces an off-policy update for flow-policy reinforcement learning

PRISM turns four real videos into 256 counterfactual variants for humanoid robot training

ABC robotics stack released with a 130K-plus-episode teleoperation dataset

The prospect of large-scale real-to-simulation systems

The case for language models at the core of physical AI

Featured in

EyeRobot 2.0 uses active gaze instead of wrist cameras for robot manipulation

QF3 introduces an off-policy update for flow-policy reinforcement learning

PRISM turns four real videos into 256 counterfactual variants for humanoid robot training

ABC robotics stack released with a 130K-plus-episode teleoperation dataset

The prospect of large-scale real-to-simulation systems

The case for language models at the core of physical AI