MIT PhD candidate Maria Bauza presents research on precise robot pick-and-place for novel objects at the UW Robotics Colloquium, introduced by Dieter Fox of the Paul G. Allen School.
Bauza contrasts flexible human manipulation with industrial robots that are often configured for a single task. Even small changes can make those systems difficult or uneconomical to adapt. Her goal is a robot that can handle varied tasks while retaining precision, with furniture assembly as a motivating example.
The focus is pick-and-place of unfamiliar objects into tight target configurations. A useful initial grasp must hold the object securely, expose features that help estimate its pose, and make placement easier. Choosing that grasp with the final task in mind can reduce or avoid regrasps.
The system trains its perception components from simulated object models and contact images. A neural network learns whether tactile images correspond to nearby contact poses, then matches real sensor input against simulated examples to produce a distribution of plausible poses. This matters because distinct poses can create indistinguishable tactile impressions; some reported examples had 2.45 millimeters of median error, while ambiguous contacts remained uncertain.
Bauza applies a similar image matching method to depth-based vision, then combines visual and tactile pose distributions. The full system selects grasps for stability, dexterity, and useful sensing, and uses simulated components to pick and place previously unseen objects. It reports up to 95% accuracy on some edges and can also work with noisier reconstructed object models.
The current system commits to a pose estimate and plan, so Bauza identifies reacting to disturbances as an important next step. She also raises whether object-independent models can retain enough precision, and how much model detail a task actually requires. Future directions include active model building and using simulated forces to guide assembly.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Maria Bauza, an MIT mechanical engineering PhD student, discusses tactile sensing for dexterous robot manipulation in a seminar hosted by the AI Robotics Seminar at the University of Toronto.
