|
Laetoli |
|
Machine learning applied to physical character animation. Artificial evolution of neural networks produced realtime controllers able to walk short distances. Although controllers were evolved using a single morphology, they were still useful when applied to very different characters. For example, controllers that learned against a 50th percentile height and weight female were found to still be useful when applied, without further evolution or training, to a 90th percentile height and weight male character. |
Performance Evaluation in Laparoscopic Surgery | |
| Automated performance evaluation of laparoscopic surgery currently requires the surgeon to operate in a simulated VR-style environment. These environments lack fidelity, especially in terms of haptic senses, during the simulation of physically complex tasks, such as suturing. By placing miniaturized positional sensors directly inside working laparoscopic tools, we allow precise tracking of tool motion, potentially even in vivo, through the full range of surgical tasks. Stastical and evolutionary machine learning techniques automatically classify motions into subtasks, which can then be compared to the motions of experts performing the same subtasks. Our approach greatly increases both the range of evaluative tasks, and the accuracy of the evaluation. |
|
Virtual Environment for Treating PTSD |
|
A "serious game" to assist the clinical treatment of post-traumatic stress disorder in US-Iraq War veterans. A Virtual Reality Exposure Therapy Application for Iraq War Post Traumatic Stress Disorder, published in IEEE VR 2006. The deployed system was featured in a CNN segement. |
|
Random-Walk Aggregation |
| A project emulating (on a smaller scale) work done by Andy Lomas. Particles are emitted from a region (in this image, a small area above the peak) and then randomly walked until colliding with either another particle or the ground plane. Various global and local forces can be added to change the resulting shape. |
|
Simple Physics |
| Simple physics simulator with a variety of numerical integrators, including RK4, verlet, forward euler and symplectic euler. Collision resitution uses penalty forces. Friction is simulated by tracking the entry point by back-projecting the velocity when collision is first detected. A zero-length spring rooted at the entry point is attached to the particle for as long as the particle remains in a collision state. Sliding friction is modeled by checking at each update if the entry-point is outside a friction cone (defined with the apex at the particle's current position and extending in the direction of the collision normal; the cone angle can be varied to simulate different materials). If the entry-point is outside the cone, it is reseated to the closest point on the circle of intersection between the collision cone and the ground. Class project for CS274c. (MPEG-2 video) |