Load-Cell-Based Bin Picking
A bin-picking system that uses load cells beneath a container to estimate object positions and update a probabilistic belief model. The approach investigates how physical sensor feedback can complement vision when objects are occluded or reflective.
Overview
This project explores bin picking with load cells mounted beneath a container, using changes in measured weight distribution to infer object positions when vision alone is unreliable.
Challenge
Vision-based bin picking struggles when objects are occluded, stacked, reflective, or otherwise difficult to segment reliably from camera images.
Solution
Load cell readings are fed into a probabilistic belief model that is updated as the robot interacts with the bin, using particle filtering and reinforcement learning to guide picking decisions.
Implementation
- Design of a load-cell-instrumented bin fixture
- Probabilistic belief modelling of object positions
- Particle filter implementation for state estimation
- Reinforcement learning for pick-action selection
- Interactive visualisation of the belief state
- Robot path planning for candidate picks
Technologies
Gallery
Technical diagram
Results
- Placeholder result — replace with a verified project outcome.
Limitations
- Placeholder limitation — replace with a verified constraint of the current approach.
Next steps
- Placeholder next step — replace with a verified planned extension of this work.
Related publications
Have a similar challenge?
Send us your technical requirements, drawing, or project idea and we will evaluate a practical approach together.