Marc H. Raibert

BigDog, the Rough-Terrain Quadruped Robot

Abstract Less than half the Earths landmass is accessible to existing wheeled and tracked vehicles. But people and animals using their legs can go almost anywhere. Our mission at Boston Dynamics is to develop a new breed of rough-terrain robots that capture the mobility, autonomy and speed of living creatures. Such robots will travel in outdoor terrain that is too steep, rutted, rocky, wet, muddy, and snowy for conventional vehicles. They will travel in cities and in our homes, doing chores and providing care, where steps, stairways and household clutter limit the utility of wheeled vehicles. Robots meeting these goals will have terrain sensors, sophisticated computing and power systems, advanced actuators and dynamic controls. We will give a status report on BigDog, an example of such rough-terrain robots.

Adjusting step length for rough terrain locomotion

The task of controlling step length in the context of a dynamic biped robot that actively balances itself as it runs is discussed. Three methods for controlling step length, each of which adjusts a different parameter of the running cycle, are discussed. The adjusted parameters are forward running speed, running height, and duration of ground contact. All three control methods are successful in manipulating step length in laboratory experiments, but the method that adjusted forward speed provided the widest range of step lengths with accurate control of step length. The three methods for controlling step length manipulated the dynamics of the system so the feet could be placed on the available footholds without disturbing the systems balance. An alternative approach which ignores balance for a single step, placing the foot directly on the desired foothold, and recovering balance later is described. >

Running on four legs as though they were one

Simple locomotion algorithms provide balance for machines that run on one leg. The generalization of these one-leg algorithms for control of machines with several legs is explored. The generalization is quite simple when muitilegged systems run with gaits that use the support legs one at a time. For these gaits the one-leg algorithms can be used to control multilegged running. The concept of a virtual leg is introduced to further extend the approach to gaits that use the legs in pairs, such as the trot, the pace, and the bound. These quadruped running gaits map into gaits that use one virtual leg for support at a time, for which the one-leg algorithms can provide control. This approach was used in laboratory experiments to control a quadruped machine that runs with a trotting gait.

Trotting, pacing and bounding by a quadruped robot

This paper explores the quadruped running gaits that use the legs in pairs: the trot (diagonal pairs), the pace (lateral pairs), and the bound (front and rear pairs). Rather than study these gaits in quadruped animals, we studied them in a quadruped robot. We found that each of the gaits that use the legs in pairs can be transformed into a common underlying gait, a virtual biped gait. Once transformed, a single set of control algorithms produce all three gaits, with modest parameter variations between them. The control algorithms manipulated rebound height, running speed, and body attitude, while a low-level mechanism coordinated the behavior of the legs in each pair. The approach was tested with laboratory experiments on a four-legged robot. Data are presented that show the details of the running motion for the three gaits and for transitions from one gait to another.

Autonomous navigation for BigDog

BigDog is a four legged robot with exceptional rough-terrain mobility. In this paper, we equip BigDog with a laser scanner, stereo vision system, and perception and navigation algorithms. Using these sensors and algorithms, BigDog performs autonomous navigation to goal positions in unstructured forest environments. The robot perceives obstacles, such as trees, boulders, and ground features, and steers to avoid them on its way to the goal. We describe the hardware and software implementation of the navigation system and summarize performance. During field tests in unstructured wooded terrain, BigDog reached its goal position 23 of 26 runs and traveled over 130 meters at a time without operator involvement.

Machines That Walk

This paper explores the notion that the control of dynamically stable legged systems that locomote in 3-space can be decomposed into a planar part and an extra-planar part. The planar part generates the large leg and body motions that raise and lower the legs to achieve stepping, that propel the system forward, and that maintain balance. The planar part of the control system deals only with 2D dynamics. The extra-planar part of the locomotion control system suppresses motion in those degrees of freedom that would cause deviation from the plane of motion. These degrees of freedom include roll of the body, yaw of the body, and translation perpendicular to the intended direction of travel.

Design and Implementation of a VLSI Tactile Sensing Computer

A new type of tactile sensor is presented that was designed to give a robot manipulation system information about contact between its hand and objects in the environment. We describe a device that is at once a special-purpose parallel computer and a high-resolution tactile array sen sor. The passive substrates of earlier tactile sensors have been replaced with a custom-designed very large scale in tegration (VLSI) device that performs transduction, tactile image processing, and communication. Forces are trans duced using a conductive plastic technique in conjunction with metal electrodes on the surface of an integrated cir cuit. An array of processors implemented within the inte grated circuit perform parallel two-dimensional convolu tions between programmable filtering masks and a binary tactile image. Data are then read from the array serially, so they can be transmitted to a control computer. A 6 x 3 array sensor with 1 mm2 tactile cells has been designed and tested. It is fully functional. In preparation for constructing large sensor arrays with hundreds of elements, the possibility of constructing defect tolerant tactile cells was explored. Analyses based on the Poisson model indicate that working arrays with 1,000 functional cells are possible if computing elements are rep licated within each tactile cell. Experiments on a 3 x 3 array sensor with redundant pairs of computing elements suggest that large tactile sensing arrays are within reach.

Hopping in legged systems — Modeling and simulation for the two-dimensional one-legged case

A two-dimensional one-legged hopping machine is modeled and simulated in order to better understand legged systems that hop and run. The analysis is focused on balance, dynamic stability, and resonant oscillation for the planar case. A springy leg with nonzero mass, a simple body, and an actuated hinge-type hip are incorporated in the model. Control of the model is decomposed into a vertical hopping part, a horizontal velocity part, and a body attitude part. Estimates of total system energy are used in regulating hopping height in order to initiate hopping, to maintain level hopping, to change from one hopping height to another, and to terminate hopping. Balance and control of forward velocity are explored with three algorithms. The feasibility of decomposing the control of running into a height control part, a forward velocity control part, and an attitude control part is verified by simulations.

Running with symmetry

Symmetry can simplify the control of dynamic legged sys tems. In this paper, the symmetries studied describe motion of the body and legs in terms of even and odd functions of time. A single set of equations describes symmetric running for systems with any number of legs and for a wide range of gaits. Techniques based on symmetry have been used in laboratory experiments to control machines that run on one, two, and four legs. In addition to simplifying the control of legged machines, symmetry may help us to understand legged locomotion in animals. Data from a cat trotting and galloping on a treadmill and from a human running on a track conform reasonably well to the predicted symmetries.

The LittleDog robot

LittleDog is a small four-legged robot designed for research on legged locomotion. The LittleDog platform was designed by Boston Dynamics with funding from DARPA to enable rapid advances in the state of the art of rough-terrain locomotion algorithms. In addition to providing a fleet of 12 robots with baseline software and development tools, LittleDog served as a cross-team common platform that allowed direct comparison of results across multiple research teams. Here we report the details of this robotic system.