How can robotic arms be made more robust?
Autonomous robotic manipulation requires many individual complex tasks to come together. A robot must be able to accurately identify the object it needs to move, plan how to move the object to its destination, and execute the plan accurately. Deep learning has helped researchers address many of these issues, but there are still many limitations which Professor Held attempts to solve.
In this lunch and talk, scholars learned about techniques to combat false positives in deep-learning-based object classification, reinforcement learning for manipulating deformable objects such as rope, and ways to detect and pick up transparent objects by averaging inputs from RGB and depth cameras.
More about Professor David Held:
https://www.ri.cmu.edu/ri-faculty/david-held/
https://scholar.google.com/citations?user=0QtU-NsAAAAJ&hl=en