My primary research interest was in the control of one-legged and four- legged robotic systems. During my PhD I developed the control and programmed the hardware prototype for the SAHR (Single Actuated Hopping Robot), which was the first monopod robot which could overcome uneven terrain using only a single motor as actuation, and without any knowledge of its uneven environment. I also developed a controller for a quadruped robot which only used a single actuator per leg.



So how does Model Based-Engineering (MBE) come into it?

Model-Based Engineering uses principles of abstration for the iterative development of a system, be it a robot or an aircarft. In the beginning, broad concepts are developed based on simplified models of the system to be developed. These models are abstractions of the real system, in that they capture those characterisitcs that are relevant for the development of the concepts.

Once concepts are complete, a more detailed design phase can begin. This can again be prototyped in the context of a model, now adding more detail for the system to be developed.

Going a step further, in the case of control development, the actual control algorithm can now be implemented and verified against the detailed model. At this stage, the control algorihtm can even be implemented in the target programming language, compiled, and run together with the model in a Software-In-the-Loop (SIL) configuration.

Once this level of verifiation is reached, a prototype system can be built. The control algorithm can be applied to the prototype and measurements can be made in order to further detail the models being used for control development. With these extended models, now verified against the prototype, a precise control algorithm can be developed wihout resorting to continuous tests with the hardware system. Rather integration with hardware serves as a verification measure at key points during the development. This way focus can be kept on the control development, as opposed to time-consuming trial-and-error runs with the hardware.

Depending on the system requirements, this process can be iterated with a second prototype or the final product. Since the control algorithms have been developed against verified models, the "surprises" when integrating with the final system should be minimal.



Following Model Based-Engineering (MBE) principles during my research

The dynamics of a physical robot are extremely complex. Think of the huge variety of parameters that vary, from imperfections in the floor that cause changes in friction to variations of the voltage in the power supply. For this reason, the analysis begins with a very simplified model of the robot, which however captures the main dynamics of interest.

In the beginning I studied the case of an actuated pendulum as a first model. Later, I moved on to study the robot more realistically, using a SLIP (Spring-Loaded-Inverted-Pendulum) model. The SLIP models a one-legged robot as a concentrated mass for the body (larger yellow sphere in figure at the right), a massless springy leg (red component), a massless foot (small yellow sphere). The body is joined to the leg with a hip, so the leg may rotate freely.

Although the SLIP is the simplest realistic model of the robot, it has been used extensively to correctly predict the behavior of the real one-legged robot, at least qualitatively. Also, the SLIP model may be used to assist in the study and control of multi-legged robots.

When the robot is on the ground, it's in the stance phase, while the flight phase is when its in the air. Below the various stages of the motion are shown. At the left the robot is just about to touch down, then it executes the stance pahse. The fourth snapshot shows the moment of liftoff and then the flight phase follows. During the flight phase the leg will swing forward (with a motor for example) and the robot will be ready for the next stance phase.

With this model, I developed first control conecpts. I later enriched the model by including more detailed friction model in the leg and foot, as well as the main dynamics of the electric drive at the hip.

At this point we developed a hardware protoype of a one-legged robot in the lab. Using measurements from this prototype it was possible to enrich the existing model and fine-tune values for physical parameters. Thereafter, the control development could mainly be done using the model. Final verification was then done with the hardware.

Going further, concepts for a four-legged robot were developed by using the experience of the one-legged system. A four-legged prototype was developed and the control concepts were evaluated on the the real system.

Using the principles of Model-Based Engineering, it was possible to progress much more systematically with control development, rather than with an approach based on theoretical control design and direct hardware implementation, which would most certainly mean a high degree of trial and error.