Robotics Device Control and Haptic Feedback

Flashcards covering key concepts from a lecture on robotic intent prediction, stroke rehabilitation interfaces, and haptic feedback systems.

React EMG

A project focused on device accuracy by reacting to when a user intends to change their ideas or movements, specifically at transition points.

Mask modeling prediction

A supervised learning framework that uses token embeddings in a transformer to teach a model how to fill in missing data and identify transition times between intents.

Intent ambiguity

The machine learning challenge where human subjects fail to generate consistent ground truths or signals for specific labels, making model classification difficult.

Reciprocal learning

A training process involving alternating phases where both the human and the robot model adjust to maximize signal separability and accuracy.

Augmented feedback

A technique providing extra information about a robot's internal processing, such as an LED display with $2$ bars showing the confidence level of a prediction.

Proprioception

The body's internal sense of its position and orientation in space, mediated by internal strain sensors in cells.

Piezo ion channels

Specialized cell sensors, such as the $Piezo\,2$ channel, that detect cell stretching and squishing to provide sensory signals for breathing, fullness, and limb position.

Sensory substitution

The use of an existing sense to compensate for a missing one, such as using vision to guide movement when touch or proprioception is lost.

Textile pneumatic pouch actuators

Inflatable fabric bags that provide tactile feedback, such as a poke, which can be arranged in leader-follower systems to assist with limb orientation.

Diffusion policy

A denoising strategy used in robotics that takes a video sequence and uses random noise and gradient descent to find the next state with the least entropy.

Motor torque limits

A hardware-based safety layer that ensures robot motors do not exceed specific force thresholds, preventing bodily injury to the user.

Curriculum learning

A data efficiency strategy used with stroke patients to leverage structured learning processes to fine-tune models on limited individual data samples.