|M.Sc Student||Adar Gaathon|
|Subject||Improving Stability and Reaching Absolute Destinations for|
the Conservative SLIP Model
|Department||Department of Autonomous Systems and Robotics||Supervisor||Assistant Professor Degani Amir|
Dynamic legged robots incorporate many advantages such as the potential ability to traverse intermittent, extreme, and rough terrain, which is too difficult for wheeled or tracked robots. Yet, this ability comes with significant shortcomings. For example, dynamic legged robots lose their stability and easily divert from their intended path when traversing an unknown rough terrain. To make things worse, the common solution of applying closed-loop control schemes demands a high bandwidth for data, acquired from accurate and fast sensors.
A well-known approach to deal with the shortcomings of dynamic legged robots is to use minimalistic control schemes, which use feedforward control of the leg angle at flight, where the control law is set a priori. They typically use only a few sensors, if any, at low frequency, commonly once per the robot’s stride. Our work use only two sensors, which are apex and touchdown detection sensors. Minimalist control methods are based upon the simple and commonly used Spring Loaded Inverted Pendulum model (SLIP), which consists of a point mass and a springy leg and breaks the movement of a dynamic legged robot to three basic phases of descent-stance-ascent. A common minimalistic control method is the swing leg retraction (SLR), which retracts the leg during the descent phase to increase stability. If tuned correctly, it can also lead the robot to a desired relative apex height above ground, which holds some benefits such as increased travel velocity and low ground reaction forces. However, when traversing an ongoing rough terrain, the control function of the SLR, which is the leg angular velocity, quickly become outdated. This leads to a rapid deterioration in the performance of the SLR and ultimately leads to the loss of stability. Other minimalistic control methods were shown to assist with reaching absolute height when traversing unknown rough terrain. These methods are a progress towards manipulating the robot to reach an absolute destination, which is fundamental for global path planning.
This thesis concentrates on improving both the robot’s stability and the ability to reach an absolute destination using minimal control methods while traversing rough terrain. The first part of the thesis prevents a premature failure; it offers an improvement for the SLR by updating in real-time the control function. In the second part, it presents a new minimalistic control method, the ‘Boxer Method’, which brings the robot to a desired absolute destination even under significant terrain variations.