Transcript Changing Objectives of Optimization

Changing Objectives
of Optimization
유리
Movement Research Lab
Our approach
Muscle optimization criteria
- Muscle activation
- Joint torque
- Metabolic energy expenditure
Evaluation
- kinematic data
- energy consumption
2
Issue
- No difference in kinematic data
- because only foot location was modulated, modulated reference
motion is barely changed.
Solution
-Modulating the whole reference motion
3
Modulating motion
We just modulate step location in Original
motion for balancing.
We want to change the motion more!
Method
-In trajectory optimization,
add CMA parameters and motion editor to give bigger
search space for CMA
- CMA can find suitable motion due to criterion
5
Modified Trajectory
Optimization
Original reference motion
CMA
Step length
Motion
editor
Frequency
Foot position
We add CMA parameter
-step length
-frequency
We make motion editor
where the frequency, step
length and velocity of
motion can be modulated.
Modulated reference
motion
6
Motion Sampling
Motion frequency =
𝑠𝑖𝑛𝑔𝑙𝑒 𝑠𝑡𝑒𝑝
#𝑓𝑟𝑎𝑚𝑒
Original motion frequency 𝑓𝑜𝑟𝑖
Desired motion frequency 𝑓𝑑𝑒𝑠𝑖𝑟𝑒𝑑
Sampling ratio =
𝑓𝑜𝑟𝑖
𝑓𝑑𝑒𝑠𝑖𝑟𝑒𝑑
Motion Sampling
desired
motion
Original
motion
Frequency
0.5
0.6
0.7
#frame
200
167
143
Up-sampling
slow
Sampling ratio = 0.65 / 0.5 = 1.3
Down-sampling
fast
Motion segment
Original motion
0
2
78
segment
l
Motion graph
L
*L = L1 + L2(interpolation)
R1
306
230
154
r
L1
458
382
R1
L2
532
Frequency
Original motion
Modulated motion
problems
Segment L 에서는 모션의 변화가 생겼으나,
원하던 방향이 아니다.
Frequency만 줄어들어야 하는데(천천히 걷는),
step length가 늘어났다.