Track recent robotics research publications, benchmarks, datasets, model releases, and technical papers. Browse dated, source-backed robotics company events.
Neuron AI aligns multisensory observations and memory in a shared stream that generates physical actions.
Training combines real robot data, simulation-amplified data and sensor-suit human demonstrations across five NEURA embodiments.
An eleven-minute 4NE1 laundry demonstration is planned for GTC Berlin on 20–22 October; the launch does not establish that future demonstration as completed.
· research publication · Google DeepMind · Foundation Models
Evidence notes
Google reports an 87% score on its Embodied Reasoning benchmark for physical and spatial problems.
In an internal test, Argon reconstructed a human cornhole video, generated simulation code and transferred a skill to an Apptronik Apollo humanoid that made seven consecutive shots.
The robotics developer release includes examples for simulation reconstruction, robot control, physical understanding and research reproduction.
· research publication · Odyssey · Foundation Models
Evidence notes
Odyssey reports Pro scores of 66.1 on Physics-IQ Verified video-to-video and 54.7 on image-to-video using prompt enhancement and best-of-eight selection.
The company reports first place in three of four WorldMark categories in its own evaluations.
Robot demonstrations include recovery from missed grasps and dropped objects; physical-system adaptation uses learned action decoders or policies.
Wuji Hand 2 performs pick-and-place color sorting from a moving conveyor using the right Flexiv Rizon 4S arm; the static left arm supplies another camera view.
The 500M flow-matching policy succeeds in 26 of 50 real-only trials and 43 of 50 trials after co-training with 100 real and approximately 1,500 grounded simulated trajectories.
Researchers teleoperate Hand 2 fingers with a Manus glove and map Vive-tracked wrist motion to the arm during real demonstration collection.
Two Flexiv Rizon 4S arms carry Wuji Hand 2; the right arm picks and color-sorts conveyor objects while the static left arm supplies another camera view.
The 500M flow-matching policy succeeds in 26 of 50 real-only trials and 43 of 50 trials after fully grounded co-training.
Training combines 100 real teleoperation demonstrations with approximately 1,500 simulated trajectories; tests include changed lighting, belt speed and objects.