Download presentation
Presentation is loading. Please wait.
Published byCalvin Wells Modified over 8 years ago
1
Optimization of a Dynamic Vaulting Behavior for LittleDog Alex Grubb and Nathan Ratliff ACRL 04/28/09
2
Task + Approach Traverse a large barrier using the LittleDog robot with maximum speed. We seek to optimize a parameterized dynamic behavior to vault over the barrier.
3
Parameterization Independent trajectory for front and back leg pairs. Each joint trajectory given fixed start and end positions and fixed time scale. Trajectory is perturbed by three gaussians with parameterized time position and magnitude.
4
Optimization Previous trajectory parameterization optimized using genetic algorithm Initially optimized gains as well but this led to unreasonably high gains and instability in both the simulator and real robot. Uses distance traveled as fitness function, given that time scale is fixed.
5
Preliminary Results
6
Other Learned Behaviors
8
Robustness Simulator differs substantially from the real world, leading to substantially different real world performance. Currently focusing on optimizing over a set of sampled simulators with slightly different frictional properties, etc. to maximize robustness of learned behavior.
9
Questions?
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.