Core Concepts
Estimating and predicting human intent to enhance robotic hand control.
Abstract
The article focuses on using a Haptic Glove (HG) to control a Robotic Hand (RH) for in-hand manipulation.
An estimation mechanism quantifies human motion signals to achieve the intended goal pose of the object.
A control algorithm transforms the synthesized intent to relocate the object.
An attention-based neural network predicts intent trajectory to compensate for communication delays.
Evaluation includes performance comparison against benchmark methodologies.
The study addresses teleoperation challenges, kinematic differences, and latency issues.
Methodology includes estimation, prediction, and control algorithms.
Results demonstrate accurate intent estimation and prediction for object manipulation.
The proposed system architecture consists of two subsystems: RH and HG.
The study aims for real-world deployment in critical applications.
Stats
The test-MSE in prediction of human intent is reported to yield ∼97.3−98.7% improvement of accuracy in comparison to LSTM-based benchmark.
Quotes
"The proposed methodology is evaluated across objects of different shapes, mass, and materials."
"The time delay between the feedback at the HG and visual feedback of the RH is counter-intuitive to a teleoperator."