Loading activity…
@stepjamuk
CEO @Neuracore_AI | Assistant Professor @imperialcollege | ex-Director of Dyson Robot Learning Lab | Postdoc @UCBerkeley w/ @pabbeel | PhD ICL w/ @ajdDavison
AI
Neuracore says robot-learning teams spend too much time rebuilding infrastructure and not enough teaching robots new skills.
AI
A user highlights a task in which a humanoid robot walks while its back is tapped and stops when the tapping stops. The user says the task uses no visual signal.
XAI
A user describes GLIDE as using an LLM to predict where human robot operators will fail, then write and revise a filter between their commands and the robot.
XAI
A post claims the system can adapt to changed tasks, objects and scenes without retraining, and transfer skills to another dual-arm robot without new demonstrations.
AI
A post about work by BAAI and collaborators describes DIDO’s approach: teach a one-step video model to retain the gripper and object details that can get lost when denoising is cut short.
AI
The team says it scaled model size and pre-training compute, then tested on one complex manipulation task in real deployment.
AI
The method guides the robot toward possible contact points, then gradually reduces that guidance during training so it can focus on completing the task, according to the post.
AI
Talk at Automate Further traces robot learning from imitation to reinforcement methods.
