Cheetah is a quadruped robot developed at the EPFL, featuring three-segment pantographic legs with passive compliant knee joints. It is used as a testing platform and a toolkit to study CPG algorithms and locomotion principles.
In this semester project, we designed a 6-axis force-torque sensor (based on strain gauges) to be integrated in each leg of the robot.
The project based itself on a previous work aimed at integrating such a sensor on another robot : the Roombot. During the previous project, the experimental results did not match the simulations regarding the sensitivity of the sensor. We therefore started by analyzing the possible causes of that issue and found a problem with the isolation of the gauges on the prototype used (for the last project).
In the design of this sensor, we focused on measuring only one force/torque component per bridge. The simulation results showed that we were successful in separating the components measured by each bridge. The estimated sensitivities seem well sufficient to detect the range of forces the robot will have to deal with.
Picture of Cheetah-robot and schematics of the position of the sensor on the leg (sensor in B, contact between the floor and the foot in A)
Design of the sensor
The new sensor is composed of three parts for an easier installation of the deformation gauges.
Picture of the sensor’s parts to be glued together
Picture of different forces applied and the gauges (in small red rectangles) designed to measure the corresponding deformation
With different torques applied
The deformations along each gauge’s axis were simulated with Solidworks, here are the results in mV (with a 5V-supply Wheatstone bridge) given unit forces/torques.
We can see that each bridge (from B1 to B6) measures almost only the force/moment it is designed for. We can notice that the bridges B4 and B5 aimed to measure moments Mx and My also measure forces Fy and Fx that create such moments.
Partial experimental results
Some tests were carried on with half-bridges on the bridges 1 and 6 (to measure Fx and Fz) :
Results give an error inferior to 7% for Fx and inferior to 5% for Mz.