Direction of Slip Detection for Adaptive Grasp Force Control with a Dexterous Robotic Hand

Abstract

A novel method of tactile communication among human-robot and robot-robot collaborative teams is developed for the purpose of adaptive grasp control of dexterous robotic hands. Neural networks are applied to the problem of classifying the direction objects slide against different tactile fingertip sensors in real-time. This ability to classify the direction that an object slides in a dexterous robotic hand was used for adaptive grasp synergy control to afford context dependent robotic reflexes in response to the direction of grasped object slip. Case studies with robot-robot and human-robot collaborative teams successfully demonstrated the feasibility; when object slip in the direction of gravity (towards the ground) was detected, the dexterous hand increased the grasp force to prevent dropping the object. When a human or robot applied an upward force to cause the grasped object to slip upward, the dexterous hand was programmed to release the object into the hand of the other team member. This method of adaptive grasp control using direction of slip detection can improve the efficiency of human-robot and robot-robot teams. I. INTRODUCTIONShortly after the advent of the artificial robotic hand, preventing grasped objects from being inadvertently dropped has been a priority [1]. This is a very important problem in the area of prosthetic hands, as limb-absent people do not have a direct sense of the grasp force applied

Learning Spatio Temporal Tactile Features with a ConvLSTM for the Direction Of Slip Detection

“…In addition, we utilise a pure tactile sensor with just 24 sensing points rather than a camera-based sensor with a 30 × 30 image resolution. In these terms, our work is also close to that of [15] because we aim to solve a similar problem using the same tactile sensor. Nevertheless, we overcome their limitations by covering more slip states, including rotational slippage (both clockwise and anti-clockwise).…”

“…In the recent work described in Ref. [15], six slip types were learnt using a neural network. The authors specifically worked on the task of learning to classify slippage in four directions (west, east, north, south) and two touching states (touch, no touch).…”

Armband with Soft Robotic Actuators and Vibrotactile Stimulators for Bimodal Haptic Feedback from a Dexterous Artificial Hand

“…In future stages of this research, we plan to integrate the armband with a direction of slip detection algorithm [25]; incorporate an algorithm for correction of the nonlinear relationship between applied air pressure and tactile sensation; conduct further psychophysics experiments in context of dual tasks developed to use simultaneously the information from the vibrotactile and pneumatic stimulators; before moving the haptic interface into life-like tests of grasp control with a dexterous artificial hand [26], [27]. Overview of the bimodal haptic actuator embedded in a close-loop system, with a subject (A) controlling the robotic hand (B-C), eliciting tactile information in the course of its interaction with objects and surfaces; information that is returned by the haptic controller (D) to a bimodal haptic actuator (E) wrapped around the subject's arm.…”

“…The second function of this controller concerns the sequential order of activation for the 3 or 5 vibrotactile stimulators. For future real-time control experiments, an Artificial Neural Network (ANN) will be built into the controller and used to detect and classify the direction of sliding contact [25]. When a sliding contact or slip occurs with the dexterous artificial hand in one direction, the controller will sequentially trigger successive vibrotactile stimulations in adjacent stimulators in a clockwise manner, radially around the circumference of the limb outfitted with the haptic armband.…”

“…1) and secured in place with a flexible elbow support brace (BRACCO S.p.A., Milan, Italy). Next, the human subjects were trained to recognize six air pressure levels (P SA = 10,15,20,25,30,and 35 KPa). These values translate into haptic forces (see Fig.…”