Acoustic Soft Tactile Skin (AST Skin)

Abstract

This paper presents a novel soft tactile skin (STS) technology operating with sound waves. In this innovative approach, the sound waves generated by a speaker travel in channels embedded in a soft membrane and get modulated due to a deformation of the channel when pressed by an external force and received by a microphone at the end of the channel. The sensor leverages regression and classification methods for estimating the normal force and its contact location. Our sensor can be affixed to any robot part, e.g., end effectors or arm. We tested several regression and classifier methods to learn the relation between sound wave modulation, the applied force, and its location, respectively and picked the best-performing models for force and location predictions. Our novel tactile sensor yields 93% of the force estimation within 1.5 N tolerances for a range of 0-30+1 N and estimates contact locations with over 96% accuracy. We also demonstrated the performance of STS technology for a real-time gripping force control application.

Figures (6)

• Figure 1: AST Overview (a): A force applied to the AST surface deforms the acoustic channels beneath the sensing surface. These channels contain reference sound waves that travel from the speaker to the microphone. The sound wave amplitude is modulated in proportion to the deformation. We use FFT to transform the modulated waves to a frequency domain and machine learning methods to find the correlation between the frequencies and (i) the contact normal force and (ii) the contact location resulting from the deformation. (b) Sensor calibration: the xARM robot is used to apply a force to the AST at a set location through a load cell with a peg attached to the robot's wrist. The dataset contains the FFT signal, the load cell force, and the contact location. ML models are trained with this dataset to predict the contact force and location. (c) We apply this novel sensor to a grip force control pick-and-place task where the system grasps objects from the YCB object set calli2015benchmarking (a soft strawberry in this example), adapts grip force to reach a target force, and then moves the object to a target location. We use the Franka Emika for this task and an SMC LEZH gripper.
• Figure 2: AST configurations: (a) AST 1 has a single channel; Double Channels: (b) AST 2a; (c) AST 2b; (d) AST 3a; (e) AST 3b; (f) AST 4a; (g) AST 4b; (h) AST 4c; (i) AST 4d; Calibration points selected for initial testing: $A = \{17,50\}$, $B= \{17,30\}$, and $C = \{17,10\}$ (S1 and S2 refer to two speakers)
• Figure 3: Variation of FFT data at locations A, B, and C when force varies from 0 to 30$^{+1}$ N
• Figure 4: Performance of different configurations of STS: percentage of force estimated with $\pm$ 0.50 N (a), $\pm$ 1 N (b), and $\pm$ 1.50 N (c) tolerance.
• Figure 5: Frame-less AST: (a) Frame-less AST (f-AST) with calibration points: $A = \{15,35\}$, $B= \{15,25\}$, and $C = \{15,15\}$ (b) skin mounted on the curved surface of a finger; (c); (d) Percentage of force estimations with $\pm$ 0.50 N, $\pm$ 1 N, and $\pm$ 1.50 N tolerance; (d) Contact location true predictions.