Contact-Aware Manipulation

Exploring inverse kinematics and whole-body strategies that respond to contact along a robot's body.

Learning to use body contact

Most manipulation systems plan around the gripper and treat other contact as a collision. My current research asks when contact at the arm, torso, or base can improve a task, and how a robot can sense and control it.

I have studied contact-aware inverse kinematics, sensing-aware placement of tactile skins, and simulated whole-body pushing, bracing, and pivoting. These are analytic and simulation studies. They establish software baselines and expose limitations; they do not demonstrate a general contact-rich policy on physical hardware.

The next research step is to connect distributed sensing to control and learning on a real robot, starting with bounded, supervised experiments.

What the current studies show

Twelve-case fixed-point contact avoidance comparison
Original fixed-weight contact-aware IK study: the avoidance objective reduced violation in all 12 numerical cases, with a nonzero residual in 11. This is an analytic result, not robot hardware validation or a result from the later ContactIK-MPC work.
Planar simulated robot and contact funnel
Planar whole-body contact simulation used to diagnose the pushing and bracing task setup. It does not establish a learned manipulation policy.