Joran Booth, Ph.D. in Mech. Eng.

Tensegrity robots are a class of robot originally imagined at NASA in the early 2010’s as a robust platform for extraplanetary exploration. They combine rigid compressive elements with flexible tensile elements, giving them superior survivability.

Published in

• Soft Robotics (2021)
• Science Robotics (2018)

Specifications

This experimental design had several advantages

• All control hardware moved on-board (all prior fluidic tensegrities and most soft robots have off-board control hardware)
• In-situ 24 strain sensor network capable of detecting the robot shape in real-time (first of its kind among tensegrities)
• Neural network with 99% accuracy for detecting direction of gravity using strain sensor data alone
• Contact sensors for detecting collisions, 12 in all

The design also had several disadvantages

• Insufficient actuator displacement, approx. 24% strain
• Insufficient rest-state tension due to the high mass
• High mass, approximately 2kg.

The disadvantages led to the robot only being able to performed controlled rolling manuevers on a downward slope.

Despite the disadvantages, the robot was successful at demonstrating novel on-board sensing and closed-loop control

Design and fabrication

This particular robot was built entirely from scratch by me, Dylan Shah, and Olivier Cyr-Chronier. Significant technical advice was provided by Michelle Yuen and Jennifer Case. I was primarily responsible for the design of the majority of the robot, including

• McKibben pneumatic actuators

• Separable triangular skins and their connection points

• Internal rod structure and end caps
• Pneumatic distribution system
  • Distributed, valve manifold system
    • i2c communication system and wiring harness
    • Logic controller pseudocode (50/50 effort with Dylan)
  • Plumbing configuration
  • Tether design

• Estensible elastomer capacitive sensors
  • Five intercalcated graphite composite layers (external layers are grounded, reducing EM interference)
  • Same-side conditioning board interface
  • i2c communication system and wiring harness

• Basic control algorithms
  • Motion primitives discovered using
    • 3D kinematic model of the robot based on a least-squares numerical solution (algorithm and initial pseudocode suggested by Jennifer Case)
    • Genetic algorithm simulation of the tensegrity

Demonstration