Watch A Computer Learn How To Walk

Wednesday, January 22, 2014

 Computer Graphics
A team of researchers has developed a realistic walking simulator for a variety of bipedal creatures. In the simulator, two-legged computer-based creatures walk in various conditions with a system using discrete muscle control parameters.

Agroup of researchers from Utrecht University and the University of British Columbia have created a  a muscle-based control method for simulated two-legged computer-based creatures where the muscle control parameters are optimized.

Through an evolutionary algorithm, the system yields effective gaits for the creatures for various parameters, including speed rotation, and even gravity.

Watch A Computer Learn How To Walk
Image Source - Geijtenbeek, van de Panne and van der Stappen

The generic locomotion control method developed titled, Flexible Muscle-Based Locomotion for Bipedal Creatures, supports a variety of bipedal creatures. All actuation forces are the result of 3D simulated muscles, and a model of neural delay is included for all feedback paths. 

The researchers' conntrollers generate torque patterns that incorporate biomechanical constraints. The synthesized controllers find different gaits based on target speed, can cope with uneven terrain and external elements, like blocks being thrown at the creatures.

Muscle Path
An example muscle path from the research.  Image Source- Geijtenbeek, van de Panne and van der Stappen
The current method still has limitations, so it won't be powering up any humanoid robots soon. Compared to the results of other studies, the walking and running motions in the system are of somewhat lesser fidelity, especially for the upper-body.

Related articles
This can be partially explained by the absence of specific arm features in the researchers' humanoid models. For now, they favored using a generic approach, but the researchers say focusing on a more faithful human
gait could make their models even more realistic.

Despite this, the team's lower-body walking motions are very close to their state-of-the-art result. "We witness a similar near-passive knee usage during swing, as well as a natural build-up of the ankle plantarflexion moment during stance," they write.

Work on an improved set of authoring tools remains an important direction for future development. Such efforts which could be further improved include: greater fidelity for the modeling joints such as the knees, ankles, and shoulders; more accurate muscle path wrapping models that interact with the skeleton geometry; giving further thought to the detail with which the target feature trajectories need to be modeled; the addition of anticipatory feed-forward control to the architecture; and the use of alternate dynamics simulators.

SOURCE  ACM Transactions on Graphics

By 33rd SquareSubscribe to 33rd Square