Facilitating Model-Based Control Through Software-Hardware Co-Design

Sidney NiBo
3 min readJan 21, 2021

This is a summary of my takeaways and further lines of inquiry after reading Facilitating Model-Based Control through Software-Hardware Co-Design authored by Ramos, Karz, Chuah and Kim.

The paper details the design process for a bipedal robotic system and the control theory behind it. It appears the team is putting forth the sentiment that by designing the controls as hardware in lockstep you can improve performance.

This is a very limited evaluation as the paper goes way over my head in large part, but hopefully, throughout the process of writing these summaries, I’ll get a better grasp of things.

Legged robotics have numerous applications, the world is designed for humans to traverse and not for wheels, so implementing bipedal locomotion is an important step in allowing robots to interact and move throughout the world without an infrastructure uphaul.

The paper brings to attention a few things of significance in regards to the hardware and controls of legged robotics. On the hardware size, fast response times, reliable torque control and high torque density are needed, as well as reliable sensors in order to reduce the strain on the control system to be capable of dealing with high levels of noise. On the software side, most control systems also apply some level of force control which means the hardware must be capable of giving force control with a reliable resolution. To address these problems the MIT Little HERMES team actually manufactured their own custom actuators, something several high profile robotics systems have also done including the SPOT Mini from Boston Dynamics and the MIT Mini Cheetah.

One takeaway from the paper was that the minutia of the system are negligible. The model for the Little HERMES was a single rigid body on massless legs which served exclusively as force sources. To paraphrase the paper, so long as the significant dynamic characteristics are not fundamentally altered you can simplify the model. This allows for the creation of more understandable control systems.

Speaking of Controls, the Little HERMES maintained balance using an Inverted Pendulum with a Reaction Wheel like system. This section I largely did not understand, however, it appears that the system modifies the magnitude and location of the reaction forces and CoP in order to ensure that the system remains dynamically balanced.

A few other interesting things I read:

The custom actuators served as electric springs recirculating the energy when negative work was done on them, similar to regenerative braking in electric skateboards and Teslas.

Having low mass limbs is very influential in improving force control range and agility. It is for this reason that the Little HERMES, and I suspect the Mini Cheetah and SPOT Mini as well, have their knee actuators stationed in the main body and belt-driven.

The Little HERMES had custom piezoelectric sensors on the feet that the team trained with a Least Squares Artificial Neural Network to perform regression and approximate the force that the foot was experiencing from the ground.

Hopefully, this was of some use. Thanks for reading and let me know of your thoughts and please correct me if I was wrong on anything.

--

--