Home Research Publications Resume Links

Modeling and Experiments with Legged Robots
Part of my research in legged locomotion focuses on dynamically stable running behaviors of actively balanced multi-legged robots. Experimental validation of controllers for running is the ultimate goal of this work. In the past, Scout II, an untethered four-legged running robot with only one actuator per compliant leg, offered a unique platform for experimentally evaluating the performance of empirically derived control laws. The fundamental contribution of Scout II has been the demonstration that fast and stable dynamic quadrupedal locomotion can be achieved without a high power leg actuator, directly changing the energy stored in the compliant legs during stance. Beyond the fundamental insights into the control and dynamics of legged locomotion, the practical implications of this fact are significant. With reliability and cost being some of the main obstacles to the commercial viability of legged robots, mechanical simplicity is essential. Aiming at autonomy and increased power efficiency, the core of our approach is to find simple control laws that properly excite the natural compliant dynamics of the system.
Scout II Sagittal plane model for Scout II
Scout II achieves dynamically stable running of up to 1.3 m/s on flat ground via a bounding gait. Surprisingly, the controller requires no task-level or body-state feedback, leading to the question why a task so apparently complex as running can be stabilized via simple control laws? To answer this question, the properties of the passive dynamics of Scout II have been analyzed using a simplified model. Numerical studies of the corresponding return map show that passive generation of cyclic motion is possible. Most strikingly, local stability analysis reveals that the dynamics of the open loop passive system alone can confer stability of the motion! These results contribute to the increasing evidence that apparently complex dynamically dexterous tasks may be controlled via simple control laws, and can be used in developing a general control methodology for legged robots, resulting from the synthesis of feed-forward and feedback models that take advantage of the mechanical system.
In addition, part of my work in this area addresses general modeling issues for dynamically stable legged robots. Although many models for open or closed loop dynamic legged locomotion have been studied, and many simulation techniques have been proposed for efficiently integrating the dynamic equations and visualizing the resulting motion, no models exist in the literature which are experimentally validated down to the actuator torque level. Yet, without such validation, the relevance of theoretical and simulation results to physical legged robots remains uncertain. In this work, the need for including motor saturation and non-rigid torque transmission characteristics in simulation models is demonstrated through an extensive suite of experimental results that documents the robot's performance and the proposed models' validity. Similar issues are likely to be important in other dynamically stable legged robots as well.

Representative publications