Human Motor Control

Using Whole-Body Control Models To Predict Human Motion

Synthesizing Human Motion: A whole-body operational space controller tracks marker positions on to a simulated musculoskeletal model.

Our Methods

We develop operational space controllers to make highly redundant human models follow an actual human's motion. Human motions are usually recorded with optical markers. They are reconstructed in simulation by controlling virtual markers attached to simulated musculoskeletal models. Simulated models are ether subject-specific (unique for each subject) or are fitted to humans. Using control allows our methods to readily scale to arbitrarily complex articulated body musculoskeletal models.

Operational Space Control: Using control, it is feasible to synthesize motion patterns for arbitrarily complex models. The model above has twenty eight degrees of freedom (including the legs). The motion it performs, however, is a deterministic function of only three control point (xyz) trajectories: one on the chin, and one each at the hands.

Our Research Goals

The key insight in designing motor neuroscience experiments is that experiments must probe how the brain resolves the under-constrained inversion of low dimensional task specifications into high dimensional actuator commands. Moving a hand, for instance, involves specifying a three translation variables, which must be mapped into control signals for fifty muscles. Lacking a formal theory to probe this muscular coordination map, present neuroscience experiments rely on brute-force designs with incremental trial-and-error based updates. These experiments are tedious, error-prone, and limited in scope.

Our human biomechanical model controllers can predict what real-world motions induce large gradients in the the neuromuscular actuation space, and HFI allows us to perform these potentially complex motions in a structured manner. These experimental methods set the stage for experiments to map human motor coordination.

Publications

[2016] Menon. S, Migimatsu. T, Khatib. O, “A Parameterized Family of Anatomically Accurate Human Upper-Body Musculoskeletal Models for Dynamic Simulation & Control, Proceedings of the 16th IEEE-RAS International Conference on Humanoid Robots, 2016. (In press)

Bibtex

[2016] Klingbeil. E, Menon. S, Khatib. O, “Experimental Analysis of Human Control Strategies in Contact Manipulation Tasks, Experimental Robotics, Springer Tracts in Advanced Robotics, Vol. xx, pages xx-xx, 2016. (In press)

Bibtex

[2014] Klingbeil. E, Menon. S, Go. K, Khatib. O, “Using Haptics To Identify Human Contact Control Strategies For Six Degree-of-Freedom Tasks”, World Haptics Sympoisium, 2014. [IEEE]

Bibtex

[2011] Menon. S, Khatib. O, “Controlling Biomechanical Models To Move Like Human Do”, (Abstract; Invited Talk), Biomechanical Engineering Conference at Stanford, 2011

Bibtex

[2010] Demircan. E, Besier. T, Menon. S, Khatib. O, “Human Motion Reconstruction And Synthesis of Human Skill”, Advances in Robot Kinematics: Motion in Man and Machine, edited by Jardan Lenarcic and Michael M. stanisic, Springer Netherlands, pages 283–292, 2010, [Springer]

Bibtex