Simulation of point to point reaching using a two-link human arm model controlled by an iLQR controller
The following articles were referred to complete this project.
Model and task
- The model was generated based on the parameters used in (1), excluding the muscle models and controlling joint torques directly.
- Four targets were placed at different locations on the 2D plane and reaching to these targets from a common starting target were simulated.
Initialization of control
The initialization of control actions at each time step was obtained by modeling the problem as an infinite horizon LQR. The reaching trajectories due to this control are unsatisfactory as shown in the initial animation and figure below (red trajectories).
ILQR controller
The controller actions were optimized iteratively using Differential Dynamic Programming as described in (2). The final control solutions shown in the figure below (blue trajectories) and animated in final animation were reached under 10 iterations for all targets. Trajectories from selected inbetween iterations are shown in grey (dark to light in increasing iteration number).
