Sree Prasanna Rajagopal

My research motivation is building bio-inspired cobots. Particularly, I’m interested in motion and manipulation design with soft actuators for semi-autonomous, human-assisted robots.


July 2013 to June 2014

Bio-mimetic pneumatic wrist prosthesis

worked with:

S.K. Dwivedy, Indian Institute of Technology, Guwahati




We show the impact of using Pneumatic Artificial Muscles (PAMs) for a prosthetic application. Our design includes a bio-mimetic wrist joint with 2 degrees of freedom, actuation using PAMs, and PID-controlled solenoid valves for pressure control. Our experimental results show a quadratic pressure-length characteristic for the actuators.




Design philosophy:

Our aim was to make a wrist mechanism as close and faithful to the human wrist as possible. In addition to mimicking the human wrist motion, we also wanted the mechanism to be as safe as a human wrist. Safety comes from the soft robotic actuators used here which tend to be more pliant. This gives room for the mechanism to be compliant and thus augment a human in a safe and complete manner.

Motion Range:

We aimed at a motion range of 130 degrees for hyperextension/hyperflexion aka. Tilt, and 60 degrees for radial flexion and ulnar flexion aka. Yaw.


Our choice of actuator was a soft pneumatic muscle. This was in-line with our mission to keep safety as a primary goal of the mechanism.

Cost vs Time:

For actuation, we had to control air flow through the muscles. Higher the flow, more force the muscle exerted on the endpoint. By controlling the flow, we could create a setpoint force. Restricted budget meant improvised. A proportional control valve ($1500) made the control easy; set voltage proportionally changed flow. We chose a cheaper option of on/off valves ($150). By using switching mechanisms such as PWM, an on/off valve could in theory simulate a proportional valve. In practice, some operating regions are unavailable. By controlling the forces on the three muscles, the wrist mechanism can reach orientations of tilt and yaw, very close to the human wrist.

Control Scheme:

With our actuators and sensors selected, we needed a control scheme for the overall mechanism. We chose to use a feedforward model with the known muscle model in combination with a PID feedback to model the unknowns.

Modeling known unknowns:

We modelled the feedforward model of the muscle as a function of load, muscle length, and the flow.

Experimental modeling:

To gather data in a continual manner, we added an IR sensor for distance measurement in our test setup. This allowed us to rapidly collect data over the three variables.

Final result:

The results were promising for the model over a range of loads from 2.5-5 kilograms. The calibration allows us to further model the PID controller with tighter values since the feedforward model is uniform over a variety of loads. Our results are published here.

tags: prosthesis