This project was about the design, manufacturing and characterization of a 3-axis force sensor for the CheetahCub, a running quadruped. Its aim is to measure the ground reaction forces. This can be the basis of a sensory feedback.
Made up of three load cells, the sensor is placed at the hip level. The development steps of the system are detailed, together with a set of design principles, specific to 3-dimensional force sensors.
The characterization is presented: among others, sensitivity, linearity, accuracy and temperature dependency were measured. The sensor’s main asset is a very high independancy between the axes (crosstalk < 1%).
A set of experiments were performed to show that the sensor can be used on a running robot, but which weaknesses have to be considered.
The three load cells: two are measuring vertical forces (side) and one is for horizontal forces (bottom).
One vertical load cell.
The system is holding a servomotor: the leg’s actuator. By this mean, external forces on the leg (i.e ground reaction) can be measured.
Flexor reflex implementation, based on the designed sensor. This could prevent the robot from stumbling. When it hits an obstalce while walking, this event can be measured by the sensor at the hip level. Here, the controller makes the robot lift the leg quickly when it occurs.
This reflex should obviously not be triggered when the robot makes a normal step.
The video shows that these two events are distinctly observed by the sensor, whatever the leg’s position.
In slow motion
A no-force controller, also based on the same sensor (at the hip level). The controller integrates the external vertical forces to set the knee angle.
Position follower on servomotor
This controller uses the torque measured by the sensor to set the servomotor angle.
As you can hear, the motor is actuating the pulley as needed, to reach the requested position.
There is NO BACKDRIVE, the position is completely constraint by the controller. Only the measured forces are used to determine the desired angle.
We could successfully detect slipping out of the sensor’s signal. As soon as the foot looses adhesion, the sensor’s signal systematically shows important noise of high frequency. Slipping detection could be used to tune the gait parameters of the robot, according to the ground type.
Slow motion footage of the experiment
Sensor’s measurments over time, during a similar experiment. Slipping occurs at t = 3.2 s. The signal becomes suddenly very noisy.