Printable Version of this PageHome PageRecent ChangesSearchSign In
Tag:
My proposed thesis is a human motion simulation that is a mix of artificial intelligence, physics, and kinesiology. The project is based on previous work that generated human motion sequences by shuffling known sequences.[1] Machine learning algorithms were used to select the most likely joint position at each time. The full body posture was determined from the combination of all joint positions.

Unfortunately, the generated motions were prone to abrupt transitions between postures and a combinatorial search space. The idea behind this thesis is to apply concepts from physics and kinesiology to the the generation of the motion sequences. The rationale for this approach is that selecting joint positions probabilistically with no other information does not consider any reason for why that position should be selected. The laws of motion and the structure of the human body could be used to determine underlying motivations to movements. The laws of motion can be used to analyze any object in motion, including the human body. Given horizontal and vertical forces and mass, we can calculate acceleration. We should also be able to calculate an objectís center of gravity, and ultimately, the degree to which an object (person) is in balance. Kinesiology will be used to tell us the limitations of the joints and muscles, and how these elements act when subject to forces. All of this information will be considered as a foundation, upon which probabilistic algorithms will be applied.

[1] E. Bradley and J. Stuart, "Using Chaos to Generate Variations on Movement Sequences," Chaos, 8:800-807 (1998)

Last modified 10 December 2007 at 8:02 pm by RhondaHoenigman