System and method for location determination and robot

US20220168899A1 · Massachusetts Institute of Technology

Patent number
US20220168899A1
Applicant
Massachusetts Institute of Technology
Filing date
2021-11-19
CPC class
B25J9/1697
Pillar
Brain

The patent describes systems and methods for locating a target object that is partially or fully occluded using RF signals from an RFID tag combined with vision sensor data, then controlling a robot to grasp or manipulate the object via trajectory planning and reinforcement learning that fuses RF perception with visual information.

Claims

A control system configured to determine location of a tagged target object based on RF signal, determine trajectory to grasp based on RF and visual info, generate control signal for robot grasping operation, and verify grasp. Includes methods for reinforcement learning policies fusing vision and RF perception using RF location as attention mechanism.

Technical analysis

Integrates RFID-based RF localization with computer vision and deep learning (CNN) for occluded object grasping in cluttered environments. Uses RF kernel and binary masking for attention, spatio-temporal rewards in RL for policy learning, enabling decluttering or direct grasping without full visual line-of-sight.

Commercial impact

Enables reliable robotic picking in warehouses, manufacturing, or logistics where objects are hidden under piles, reducing human intervention, improving automation efficiency, and lowering costs in occluded environments.

Related patents

All patents · Source filing