Human-like reaching movement can be efficiently generated by appropriately replicating an initial ballistic segment and a concluding corrective segment. During the ballistic segment, a damped inertial plant is driven toward a fixed target using a specific form of position-based feedback, referred to as a displacement-normalized actuation program (DNAP). Instead of trying to regulate a particular kinematic profile, the DNAP is used to adjust force (or torque) during the movement. This choice of control structure causes the resulting movement to match key characteristics of human movement that are not easily replicated with traditional control architectures. Only two parameters are required to scale the DNAP template, and these values can be computed as functions of the aiming task configuration. Thus, ballistic control is effectively implemented using a lookup table that contains the DNAP template, and two scaling factors. Stochastic variability is intentionally introduced into these scaling parameters to produce kinematic and temporal randomness that is similar to that found in human subjects making repeated movements.;As the movement concludes, control is then switched to a conventional feedback architecture that terminates the motion at a desired endpoint. By making the mechanical system respond as a second-order system, the kinematic response can be modified using just two additional parameters. Again, these control values can be expressed as functions of the aiming task configuration. To merge the ballistic and corrective phases, a particular compensator state is associated with the transition instant. Since each ballistic trajectory transitions into corrective action with a different velocity and acceleration, each movement follows a unique path. Endpoint variation that is typical of aimed human movement is accomplished by adding stochastic noise to the target displacement. Simulated movements generated with the proposed method are shown to be consistent with human motion.
展开▼
