Like its predecessors, ROBART III is strictly a laboratory prototype, never intended for fielding in the real-world: 1) it is not waterproof, 2) its mobility is constrained to planar floor surfaces, so it can’t ascend or descend stairs, 3) it’s not defensively armored, 4) it’s not rugged, and, 5) it cannot self right in the event it flips over.
ROBART III is instead a concept-development platform optimally configured to support its intended research and development role in a laboratory environment. Specific research thrusts include: 1) enhanced reflexive teleoperation, 2) automated target acquisition and tracking, 3) simultaneous localization and mapping, 4) natural language understanding, and 5) augmented virtuality.
As a research platform, ROBART III is one of the most sophisticated mobile robots in the world, and has been featured numerous times on the Learning, History, and Discovery Channels. It was recently ranked number 16 in Wired Magazine's survey of the 50 best robots ever (January, 2006).
Enhanced Reflexive Teleoperation
The basic issue being addressed here is the difficulty of controlling a mobile robot equipped with a surveillance and/or targeting camera and also an articulated weapon system. Experience gained through extended use of conventional teleoperated devices of this type has revealed considerable shortcomings from a man/machine interface point of view. Simply put, if a remote operator has to master simultaneous manipulation of three different joysticks (i.e., one for drive and steering, another for camera pan and tilt, and yet a third for weapons control), the chances of hitting a moving target are minimal. In actuality, the single task of simply driving a teleoperated platform using vehicle-based video feedback is no trivial matter, and can be stressful and fatiguing even under very favorable conditions.
The innovative thrust behind the ROBART III development is an extension of ROBART II's reflexive-teleoperation (i.e, guarded motion) to incorporate sensor-assisted weapon control into the integrated package. The approach involves making two of the three controllable elements (i.e., drive control, camera control, and weapon control) slaves to the third, so the human operator only has to deal with one entity. For example, the surveillance camera can be slaved to the weapon, so that the camera looks wherever the operator points the gun. If either the weapon pan-axis controller or the camera pan-axis controller approach their respective limits of allowable travel, the robot's mobility base automatically rotates in place in the proper direction to restore the necessary range of motion. Alternatively, the weapon can be slaved to the surveillance camera, and so forth. In all cases, final closed-loop control of weapon pan-and-tilt can be provided by the video target-acquisition system.
Automated Target Acquisition and Tracking
Initial 360-degree motion detection is supported by a ring of passive infrared sensors around the neck, an AM Sensors microwave motion detector behind the face plate, and a Visual Stone omni-directional camera mounted on the head. Fused outputs from these sensors are used to cue a Canon high-resolution pan-tilt-zoom (PTZ) camera in azimuth and elevation, which further assesses the potential target. For static (i.e., motionless) targets, the PTZ protocol has been integrated with a two-stage search-and-engage algorithm, wherein the vision system first performs a wide-area scan for a pre-taught class of objects, then cues the PTZ camera to zoom in and search for specific “vulnerabilities” associated with that particular target. The non-lethal weapon is automatically trained accordingly with the aid of a bore-sighted targeting laser, and then fired under operator supervision.
Simultaneous Localization and Mapping
For increased versatility as a prototype response vehicle, ROBART III’s navigation strategy required modification to support fully autonomous operation in previously unexplored interior structures (i.e., with no a priori information or map). Starting in FY-03, the navigation and collision avoidance schemes have been significantly enhanced through technology transfer of improved algorithms developed under DARPA’s Tactical Mobile Robot (TMR) and Mobile Autonomous Robot Software (MARS) programs (Pacis & Everett, 2004). Under a Memorandum of Agreement, SSC San Diego has subsequently tasked the Idaho National Laboratory (INL) to assist in the coordinated development, evaluation, and transfer of robotics technology that mutually benefits both Department of Defense and Department of Energy missions.
As one of the key TMR contributors, SRI developed a mapping technique called Consistent Pose Estimation (CPE) that efficiently incorporates new laser scan information into a growing map. Within this framework, SRI has addressed the challenging problem of loop closure: how to optimally register laser information when the robot returns to an area previously explored. With CPE, it is possible to create high-resolution maps and repeatedly execute the accurate path following necessary for high-level deliberative behavior.
CPE is another method for performing simultaneous localization and mapping (SLAM), based on original work by Lu and Milios (1997), who showed that information from a robot’s encoders and laser sensors could be represented as a network of probabilistic constraints linking the successive poses of the robot. The encoders relate one robot pose to the next via dead-reckoning, and the laser scans are matched to give further constraints between robot poses, including constraints for when a robot returns to a previously-visited area (Gutman and Konolidge, 1999). CPE provides an efficient means of generating a near-optimal solution to the constraint network and yields high-quality metric maps. Once a map has been made, it can be used to keep the moving robot localized.
Natural Language Understanding
SSC Pacific is investigating a natural-language interface that would allow a supervised autonomous robot to be given fairly unstructured verbal direction, no different from the procedures used to instruct a human to perform the same task. For example, suppose the robot has penetrated an underground bunker and is streaming back video that shows an open doorway in the center of the far wall of the room just entered. A human monitoring this video might converse with the robot as follows: “Find the doorway in front of you.” The robot would then analyze the current video, looking for predefined scene attributes that suggest a door frame or opening, highlighting its choice with a graphic overlay, then verbally respond, “Request confirmation of selection.” If the robot’s vision system locked onto the same doorway the observer had intended, the human would acknowledge as follows: “Affirmative.”
If for some reason the robot selected the wrong door, or a set of scene attributes that was in fact not a door at all, the human would respond differently: “Negative, look to your left.” (Or right, as the case may be.) The system would then shift focus accordingly to the next set of scene attributes that looked like a doorway, again ask for confirmation, and so forth. Once the human and the robot were in sync, the human could issue additional voice prompts to influence the robot’s further interaction with the doorway. One example could be to zoom in on and perhaps even illuminate for better assessment, or to enter the doorway and continue searching on the other side.
Three 16-bit computers work together to give ROBART III its advanced autonomous capabilities: 1) the Torso Computer is the central computer system, responsible for gathering and processing sensor data from the sonars, initiating speech output, and controlling the integrated motion of the non-lethal weapon, and head; 2) the Vision Computer, located in the head, is responsible for converting live video from the various cameras into a sequence of digital images for processing; and 3) the Drive Computer, located in the Mobility Base, is responsible for controlling the drive motors, and gathering sensor data from the Sick ladar, Sharp I/R rangerfinders, KVH fiber-optic gyro, and the Microstrain compass. (In addition, multiple 8-bit microcontrollers are employed for low-level sensor processing and actuator control.) These computers communicate over an Ethernet network and use the Player protocol to share information.
The left and right drive wheels are driven by a pair of 12-volt electric wheelchair motors identical to those used on ROBART II. System power is supplied by a 80-amphour 12-volt gel-cell battery which provides for several hours of continuous operation between charges. Appropriate hardware upgrades have recently been made to support the more sophisticated navigation, collision avoidance, and mapping schemes, to include a MicroStrain gyro-stabilized magnetic compass, KVH fiber-optic rate gyro, and a Sick scanning laser rangefinder. Full-duplex data communication with the PC-based host control station is accomplished via a 9600-baud Telesystems spread-spectrum RF link.
The non-lethal-response weapon chosen for demonstration purposes is a pneumatically powered dart gun capable of firing a variety of 3/16-inch diameter projectiles. The simulated tranquilizer darts (20-gauge spring-steel wires terminated with 3/16-inch plastic balls) illustrate a potential response application involving remote firing of temporarily incapacitating rounds by military or law enforcement personnel. A rotating-barrel arrangement was incorporated to allow for multiple firings (six) with minimal mechanical complexity. (The spinning-barrel mechanism also imparts a rather sobering psychological message during system initialization.)
The darts are expelled at high velocity from their 12-inch barrels by a release of compressed air from a pressurized accumulator at the rear of the gun assembly. To minimize air loss, the solenoid-operated valve linking the gun accumulator to the active barrel is opened under computer control for precisely the amount of time required to expel the projectile. The gun accumulator is monitored by a Micro Switch 242PC150G electronic pressure transducer, and maintained at a constant pressure of 120 psi by a second solenoid valve connected to a 150-psi air source. All six darts can thus be fired in rapid succession (approximately 1.5 seconds) under highly repeatable launch conditions to ensure accurate performance. A visible-red laser sight is provided to facilitate manual as well as automatic targeting.
The pneumatically-powered non-lethal weapon is for laboratory demonstration purposes only. It supports the vision-based weapon-control research without undue risk to personnel. From a safety perspective, a local Ready/Standby switch enables the air compressor and secondary accumulator charging, and a local Arm/Safe switch physically interrupts power to the trigger solenoid valve. There are parallel software enables for both these same functions on the remote OCU. Two separate control lines are employed for the trigger solenoid, one active-high and the other active-low, to minimize chances of inadvertent activation during initialization or in the event of a computer reset. A three-way emergency override is also provided: 1) two local E-Stop buttons on the Mobility Base; 2) an RF kill pendent; and 3) a remote E-Stop button at the control station. A fiber-optic sensor on the gun is used to determine load status for each barrel. (There is also a remote accumulator dump valve that we don't typically use.)