State of the Art in Robotics
Robotics is a field that has many exciting potential applications. It is also a field in which expectations of the public often do not match current realities. Truly incredible capabilities are being sought and demonstrated in research laboratories around the world. However, achieving these capabilities with real robots in real environments faces many hurdles. It is true that robotic systems can be stronger and faster than humans, can go places too dangerous for a human to venture, and can operate without fatigue while performing highly repetitive and precise tasks. However, it is very difficult to build a mechanical device (e.g., a robotic arm) that has dexterity comparable to a human’s limbs. It is even more difficult to build a computer system that can perceive its environment, reason about the environment and the task at hand, and control a robotic arm with anything remotely approaching the capabilities of a human being.
Hollywood’s depiction of robots often endows them with human-like intelligence and decision-making capabilities, but real robots fall far short of this image. A robot is simply a machine that “synthesizes some aspect of the human function.”1 In general a robot involves some level of automation, which is the attribute of being able to perform a task or a sequence of tasks and adapt to a well-defined and predetermined class of variations. A robot may also exhibit autonomy, which is the ability to make decisions the way a human being might make decisions. However, the level of autonomy that has been achieved in today’s robotic systems is no match for even the simplest decision-making capabilities of a human.
Many robots are teleoperated. In teleoperation, a human operator controls the robot directly while monitoring some or all the information that the robot sensors acquire. Teleoperated robots have been used effectively by human operators to augment their skills or to be able to operate in remote, usually hazardous or inaccessible, environments. For example, the manipulators used on the International Space Station (ISS) and the shuttle are teleoperated. Surgical robots that allow surgeons to perform procedures while operating through tiny ports are also teleoperated. The key feature of teleoperation is that it exploits the perceptual capabilities and reasoning power of the human operator rather than relying only
on the sensors and computers available to the robot. A key requirement for successful teleoperation is that the communication link between the human operator and the robot is sufficient to provide enough information for the remote operator to make decisions and to issue appropriate control commands in a correct and timely manner. Teleoperated robots typically require and exhibit very little autonomy because of the presence of the human operator in the loop.
It is useful to look at some well-known applications of robotics to understand the difference between automated, autonomous, and teleoperated robots.
One of the most visible and successful application of robotics is in factories and on the shop floor. Here, reprogrammable, multilink robotic arms have replaced special-purpose machines to perform precise and quick repetitive operations, such as pick and place tasks, for handling parts and tools and for assembling parts. The advantage of using robots in these applications is that their reconfigurability and flexibility make it possible for one assembly line to be multifunctional and to be adapted for a range of parts or products. However, a production facility or a factory is typically a highly structured environment. Precisely manufactured parts arrive on schedule at predetermined positions and orientations for robotic operation, and all operations are, for the most part, predictable. Once a robot is programmed, very little “intelligence” or autonomy is required of the robot for it to perform its limited set of functions. Very little adaptation to uncertainties is required. In spirit, these robots are closer to machines like programmable looms or dishwashers than to Hollywood’s R2D2.
Another recent, very visible application of robotics is the pair of Mars Exploratory Rovers (MERs), Spirit and Opportunity. These very successful mobile robots exhibit multiple levels of autonomous or semi-autonomous operation. These rovers have sensors that provide information about the environment in which they are operating, about their position in that environment, and about the status of the task they are performing. The sensors provide information to computers, which reason about the state of the robot and the environment and calculate the commands sent to the robot’s actuators to control its motion and activities. Some of this reasoning is done onboard the vehicle. However, much of the high-level reasoning and decision making is done by the remote human users, albeit infrequently because of the time delays associated with communication between the rovers and mission control on Earth. For example, remote human users set the science objectives (e.g., on which rock to place an instrument) and issue high-level commands (e.g., “go to that rock”). The rovers then execute these commands using onboard sensors and computers to determine and follow safe paths through the terrain. Importantly, the onboard autonomy is limited primarily to the specific tasks of navigation and instrumentation placement. The rovers have some limited ability to adapt to operating conditions and the environment. When unexpected situations or failures are encountered, the rovers can stop and wait for the remote human users to issue a new set of commands. Human users can also make the decision to send new software to the rovers or patch software bugs that may be discovered during the mission. Thus, while these robots are not, strictly speaking, teleoperated, there is an element of teleoperation in the functioning of these rovers. At the same time, the rovers exhibit a significantly greater degree of autonomy than the automated factory robots discussed earlier. This combination of autonomy with an element of teleoperation is often called supervised autonomy.
There are many remotely operated vehicles like Spirit and Opportunity that have been deployed on Earth. Rovers have been used for nuclear reactor inspection at Three Mile Island and have been deployed by the military for de-mining in Bosnia and for reconnaissance in caves in Afghanistan. In Iraq teleoperated rovers with manipulators are used for disruption and disposal of improvised explosive devices. Robotic submersibles have been used in the deep sea for exploration tasks by the marine science community, for inspection and maintenance tasks by the oil industry, and for salvage of wrecks like the Titanic. The level of autonomy employed in these devices varies. It is not feasible to teleoperate
the MERs because of the time delays associated with communications; hence supervised autonomy is used. It is feasible, however, to teleoperate a vehicle driving over a minefield. Thus a military robot clearing mines through a minefield may not require the level of autonomy that the MERs require.
Robots can also be seen in the service industry. There are commercial products for vacuum cleaning, for mowing lawns, and for assisting people with disabilities. Humanoid robots are being developed for entertainment. There are many sophisticated toys that employ robotics technology. Amusement parks use programmable, articulated mechanical devices to mimic biological motion. While many of these applications provide successful examples of autonomous operation, there are no examples of dexterous manipulation.
Deciding what tasks can or should be performed autonomously by a robotic system depends heavily on the details of the specific mission. Further, enabling those autonomous operations requires an extensive, dedicated research and development program, which begins in the laboratory and culminates in field demonstrations before actual deployment on a mission. For the MER rovers, autonomous navigation was identified as having significant mission benefits and was achieved only after years of focused research and development, such as identifying obstacles using computer vision and relative state estimation using wheel, inertial, and optical sensors. Manipulation with robotic arms is a very different type of task and requires a similar, focused development activity if it is to be automated at any level. Robotic arms have been used extensively on the shuttle and on the ISS to perform assembly-class operations, but up to now all of these operations have been done in a teleoperated mode with no autonomy.2 Significant training of the astronauts is required to qualify them to use these robotic arms.
Automated rendezvous, capture, and grappling of HST and robotic servicing with dexterous manipulators cannot be performed via direct teleoperation because of the time delays in the communication link between the orbiting robot and the ground station.3 Supervised autonomy is the appropriate mode of operation for the robotic servicing mission. It allows shared control where the onboard computers can control the motion of the arms and effectors based on sensory information while human operators on the ground can make mission-critical decisions. However, the successful implementation of supervised autonomy requires that the manipulators, sensors, and control software be sufficiently sophisticated to perform assembly and disassembly tasks in an environment that is not well structured, unlike the structured environment of the factory and the shop floor, for example.
It is also important to note that although supervised autonomy has been extensively studied in research laboratories, its robustness and reliability for a mission as complex as the HST servicing have not yet been verified. There are very few examples of field-tested space operations involving manipulation or assembly with autonomy or supervised autonomy. In 1970, rendezvous and capture with a non-cooperative target were performed by the Soviets with a human operator in control and without any
communication time delays. In 1998, collaboration between ESA and NASDA produced a moderately successful demonstration on the Japanese Engineering Test Satellite (ETS) VII. This involved manipulation of a 2-meter-long, six-degree-of-freedom manipulator arm attached to a 2500 kg satellite with the coordinated control of the manipulator and the base. The ETS VII mission demonstrated autonomous rendezvous and capture of a target satellite. However, in this demonstration, the target was specially designed for capture, with appropriate fiduciaries for relative orientation, positioning, and capture. Thus the proposed HST robotic servicing mission will require the development, testing, and validation of new software and hardware, which would advance the state of the art of robotics technology.