Uncertainty-Aware Finger Force Estimation for Reliable Box Manipulation
With force adaptation
FEELWithout force adaptation
Baseline10 trials per condition, using the same box specification and predefined motion. Reliability = successful placement without recorded damage.
Applied to all experiment grids below.
Success indicates task completion without recorded damage. Failure indicates an incomplete task or recorded damage.
With force estimation experiments
(Kp, Kd) = (200, 400)Force-adaptive manipulation driven by FEEL estimates.
Without force estimation experiments
(Kp, Kd) = (200, 400)Fixed finger commands without force-estimation feedback.
Gain comparison experiments
Ten force-adaptive box-manipulation trials for each controller-gain setting.
(Kp, Kd) = (100, 200)
RMSE: 0.233 ± 0.005 N
(Kp, Kd) = (200, 400)
Used in the main experimentsRMSE: 0.251 ± 0.010 N
(Kp, Kd) = (300, 600)
RMSE: 0.255 ± 0.008 N
Data collection
Wearable F/T reference measurements synchronized with per-finger proprioceptive histories.
Approximately 25,000 synchronized estimator samples and an additional 5,000 calibration samples were collected per finger. Robot states and F/T measurements were aligned by timestamp.
Uncertainty-aware force estimation
Per-finger force estimates and predictive uncertainty under proprioceptive ambiguity.
불확실성 기반 힘 추정 영상
손가락별 추정 힘과 예측 불확실성이 함께 보이도록