A Revolutionary Wheelchair Concept
Teaching Robots To Cooperate

A Revolutionary Wheelchair Concept

A wheelchair with revolutionary maneuvering characteristics is being developed by CyberTrax Innovative Technologies using an ORNL-designed platform. Steve Killough (left) and François Pin (not pictured) developed the algorithms and innovative engineering that make the platform work. Trying out the CyberTrax prototype, called the Transportable roving omnidirectional vehicle replacement (TransRovr), is Jenny Brewington. The invention was a Discover Magazine Award finalist. Photograph by Tom Cerniglio.

People who use powered wheelchairs enjoy freedom from strenuous physical effort and from constant dependence on caretakers. Soon, they will enjoy several degrees more of another kind of freedom. A revolutionary omnidirectional chair based on an ORNL invention will allow for a freedom of motion and maneuverability hitherto impossible in wheelchair mobility.

CyberTrax Innovative Technologies of Tampa, Florida, plans to begin manufacturing the TransRovr (transportable roving omnidirectional vehicle replacement). CyberTrax signed a licensing agreement in 1996 for the algorithms and engineering technologies patented by ORNL researchers that provide the vehicle’s extraordinary maneuverability. The invention at the heart of the wheelchair, the omnidirectional holonomic platform (OHP), evolved from a DOE project to develop mobile robots for work in hazardous environments. The OHP was a 1993 R&D 100 award winner, and the TransRovr was a 1997 Discover Magazine Award finalist.

The TransRover wheelchair, based on a
platform with spin-on-itself capability,
offers a range of motion and
maneuverability not possible
in today’s powered

The base of the battery-powered wheelchair is a platform propelled by three freely rotating spherical casters (instead of wheels) that steer around a vertical axis. Its wheels, power train, and infrastructure are all hidden within the platform.

The horizontal motion capabilities of the advanced wheelchair allow unrestricted, resistance-free, 360-degree movement. The chair freely spins in a circle, moves sideways, zigzags, moves at any angle without turning, and changes directions without having to be stopped and restarted. Because the casters rotate freely, steering does not depend upon forcing the front wheels of the chair into position, as with a conventional chair. The chair turns by simply rotating on its platform base, so the wide turning radius needed for conventional wheelchairs is eliminated.

The maneuverability of the advanced wheelchair will also eliminate the wasted motion and power consumption that results from fixed-wheel steering. It allows users to choose routes based on their own preferences, rather than on the maneuvering limitations of their wheelchairs. The agility of the chair allows access to places that are awkward for conventional wheelchairs. For example, it will be able to zip into and out of tight corners without turning.

CyberTrax plans several other improvements for its TransRovr, including an improved seat design, superior ergonomics, and advanced battery technology. Improved circuitry will provide rapid internal recharging of the batteries, which will reduce downtime.

The light weight and three-piece modular construction of its design will make the wheelchair easy to pick up and transport in an automobile trunk or back seat. The cost of the advanced wheelchair is expected to be comparable to that of conventional electric wheelchairs. The TransRovr has the potential to free users from the limitations of conventional powered wheelchairs, which may eventually become obsolete.

Funding for the research that led to this invention was provided by DOE’s Office of Energy Research, Office of Basic Energy Sciences, Division of Engineering and Geosciences, Engineering Research Program.

Teaching Robots To Cooperate

Parents watching over their children at play? No, Rich Sincovec, director of ORNL’s Computer Science and Mathematics Division, and Lynne Parker, robotics researcher who developed ALLIANCE, test ORNL-developed software that can direct teams of robots like these in carrying out missions cooperatively. Photograph by Curtis Boles.

You’re helping some friends paint a room. After finishing a wall, you see that the guy painting the window trim needs help, so you grab a brush and pitch in. When the two of you are through, as the others finish painting the baseboards, you clean the brushes and wipe up the paint spots on the floor.

It seems easy enough, but that simple scenario represents a range of behaviors so complicated that science cannot explain them: perceptions, deductions, motivations, choices. No one really knows how humans make the constant adaptations necessary to work together. Imagine how difficult it is to program similar behaviors into robots so they’ll work as a team.

In the past, robotics research tended to focus on developing a single robot that could do all the required jobs on a mission. One problem with that approach is that such a Renaissance robot would need to have an enormous range of capabilities. A bigger problem is that one malfunction in it could scratch the whole mission.

ORNL is using software to “train” robots to
carry out tasks as a team, reducing the
chances that a malfunctioning robot
will cause an entire mission to fail.

ORNL is focusing instead on developing multiple, less-complex robots that can work in teams to accomplish missions. A software system called ALLIANCE, developed by an ORNL researcher, is designed to “motivate” robots to carry out a mission together and enable them to make rudimentary adaptations to the work environment.

ORNL’s robot team, four R2D2 lookalikes, has successfully demonstrated the ALLIANCE architecture in carrying out tasks such as cooperative manipulation, cooperative observation of multiple targets, and movement in formation.

ALLIANCE gives each robot the ability to select appropriate actions for itself in light of its teammates’ actions and the status of the mission. Robots fail a lot, and researchers can’t foresee and program a response for every possible failure. ALLIANCE enables the robots themselves to determine how to respond to mission-threatening failures.

“Motivational behaviors” are programmed into the individual robots to direct them in selecting actions. The primary motivations are called impatience and acquiescence: impatience drives a robot to take over a task that is not being completed by another robot; acquiescence allows a robot to give up a task it cannot complete successfully.

ALLIANCE distributes control equally to each robot team member, allowing each to select its tasks without any central command. Bandwidth limitations preclude extensive conversation or negotiation among robots to coordinate the work. Instead, each unit broadcasts information periodically to the others about what it is doing. The robots use sensory feedback to monitor their own and other robots’ performance, to indicate whether adequate progress is being made on each task.

For example, Robot A may be periodically broadcasting, “I’m painting this wall.” If Robot B’s sensors indicate that, on the contrary, Robot A is just standing there waving a brush and the wall isn’t being painted at all, Robot B will take over the task. Robot A will give it up gracefully. The task and the mission can be accomplished despite the failure of Robot A.

The potential applications for cooperative robot teams are legion: almost any task in which it is desirable to reduce human exposure in dangerous tasks or tedium in highly repetitive ones. The possibilities include hazardous waste cleanup, surveillance and monitoring in contaminated areas, industrial and building maintenance, earthmoving, mining, and military applications such as locating and detonating land mines.

During the 1960s, people thought robots would be doing housework and other disagreeable tasks by now. But it’s harder than The Jetsons led us to believe. Developing algorithms (step-by-step problem-solving procedures) to program robots for cooperative behaviors and discretionary actions is enormously time-consuming. As a next step, ORNL is working on “automatic design” of behaviors, which involves generating algorithms automatically, to reduce the programming complexity.

A goal is to develop a robot that chooses
its next move in the real world.

ALLIANCE is an early step toward building functioning robot teams that can handle real tasks. Developing a robot to play chess, with its strict rules and set scenarios, is nothing compared with developing one that can choose its next move in the real world.

The development was supported by DOE’s Office of Energy Research, Basic Energy Sciences, Engineering Research Program.

Next article