Talk title: Mechanical Impedance Compositionality Manages Many Degrees of Freedom
Abstract: Physical collaboration with humans promises new applications of robots in domestic, industrial and clinical contexts. Already robots have successfully delivered physical-contact therapy to aid recovery after stroke. However, physical interaction with humans engenders challenges. Safe physical interaction requires robust stability despite contact and coupling with unpredictable humans. Passivity at any interaction port is sufficient to guarantee that contact and coupling will not induce instability; unfortunately, passivity is compromised by null-space projection methods designed to manage robot redundancy. An alternative based on simple superposition of mechanical impedances preserves passivity but may induce task conflict. Remarkably, recent work has shown that a sufficiently large nullspace dimension eliminates those conflicts. Even if robots emulate the passive impedance of humans, the dynamic structure of human neuro-motor control limits human-robot interaction. For curved motions, humans exhibit a strong preference for a relation between path curvature and tangential speed that is consistent with maximizing smoothness. Robot motion that fails to respect this preference severely compromises human-robot interactive performance. Constrained-motion tasks (e.g. turning a crank, common in tool use) impose similar limitations. Despite visual feedback and ample time to execute corrective actions, humans are unable to suppress a correlation between limb configuration, speed fluctuations and normal forces applied to the constraint. Together these observations support an account of human motor control based on a composition of primitive dynamic actions, each action described as an equivalent network (a generalization of the classical electrical equivalent circuit) comprising motion and impedance simultaneously. The competence of this account will be reviewed.