|
Briefing
|
- Jenga-Playing Robot – MIT researchers developed robot that can play Jenga using machine learning
- Physical Components – Include retrofitted industry-standard ABB robotic arm with external camera, soft-pronged gripper, and force-sensing wrist cuff
- Machine Learning – Robot measures force exerted on block and compares these to previous moves made, whether successful or not, which then helps robot to decide whether to keep pushing block or move to new one
- Training – Trained on 300 moves instead of tens of thousands, with data on outcomes clustered to represent certain block behaviors, such as easy to move, difficult to move, or caused tower to topple down
- Potential Applications – Include robot that can separate recyclable trash and assemble consumer products
|
|
Accelerator
|
|
|
Organization
|
Massachusetts Institute of Technology
|
|
Source
|
|
|
Original Publication Date
|
January 30, 2019
|
