DSC TOC Proceedings of IMECE2002 ASME International Mechanical Engineering Congress & Exposition Proceedings of IMECE2002 November 17-22, 2002, New Orleans, Louisiana ASME International Mechanical Engineering Congress & Exposition November 17–22, 2002, New Orleans, Louisiana IMECE2002-32479 IMECE2002-32479 ROBOTIC HIGHWAY SAFETY MARKERS Shane M. Farritor Assistant Professor Department of Mechanical Engineering University of Nebraska Lincoln, NE 68588 sfarritor@unl.edu ABSTRACT Proper traffic control is critical in highway work zone safety. Traffic control devices such as signs, barricades, cones, and plastic safety barrels are often used. Accidents can occur because of improper work zone design, improper work zone housekeeping, and driver negligence. One solution is to automate safety devices. Mark E. Rentschler Research Assistant Department of Mechanical Engineering Massachusetts Institute of Technology Cambridge, MA 02139 mrentsch@mit.edu Proper traffic control is critical to work zone safety. Traffic control devices such as signs, barricades, traffic cones, and plastic safety barrels are often used. This paper presents a mobile safety barrel robot. The Robotic Safety Barrels are the first elements of a team of Robotic Safety Markers (RSM) that includes signs, cones, and possibly barricades and arrestors. To be practical the system must be reliable and have a low per robot cost. A robot that malfunctions could enter traffic and create a significant hazard. Also, multiple safety markers are used and barrels are often struck by vehicles. Safety markers with a high replacement cost are not practical. This paper describes the motivation for the robotic safety marker system and how it could improve work zone safety. The design of three robot prototypes is presented. A control architecture is discussed that has been implemented in simulation and partially tested on the prototype robots. INTRODUCTION The safety of highway construction and maintenance workers is extremely important. For every billion dollars spent on road construction, more than 33 people die in accidents related to that work [1]. Studies have found that accidents in work zones have increased by 6.8% in the some states and as much as 119% in the state of Virginia [2]. The nation’s highway system is aging and is in need of constant repair. Increased maintenance and construction will result in more injuries and more fatalities unless new ways are found to prevent these accidents. Figure 1: Robotic Safety Barrel One method to improve work zone safety is to implement a smart work zone consisting of automated devices that improve safety. This paper presents a mobile robotic safety barrel, Figure 1. Safety barrels are placed on the periphery of the work zone to guide traffic and to serve as a visible barrier between traffic and work crews. These barrels consist of a brightly 1 Copyright © 2002 by ASME colored plastic drum (approximately 130cm high and 50cm in diameter) that is attached to a heavy base. Often, hundreds of barrels are manually placed in a typical work zone. The Robotic Safety Barrel (RSB) replaces the heavy base of a typical safety barrel with a mobile robot. The mobile robot can transport the safety barrel and robots can work in teams to provide traffic control. constant presence of safety markers leads to driver complacency. In addition, deployment and retrieval activities are extremely dangerous since the construction worker is very close to passing motorists. In this country, the cost of congestion to motorists in lost productivity is estimated to be approximately $100 billion annually, not including the cost of wasted fuel and environmental damage [3]. Independent, autonomous barrel motion has several advantages. First, the barrels can self-deploy, eliminating the dangerous task of manually placing barrels in busy traffic. Second, the barrel positions can be quickly and remotely reconfigured as the work zone changes. Barrels could continuously follow work crews to maintain optimal placement for safety. Passing motorists are also partly to blame for accidents in construction zones. Typically, motorists only slow 10-mi/hr upon entering a work zone [4], therefore it is very important to make the driver aware of dangerous conditions. The Robotic Safety Barrels are the first elements of a team of Robotic Safety Markers (RSM) that includes signs, cones, and possibly barricades and arrestors. To be practical the Robotic Safety Marker System must be reliable and have a low per robot cost. A robot that malfunctions could enter traffic and create a significant hazard. The system will be designed so critical components (software and hardware) have multiple redundancy. Cost per robot is also critical. Multiple safety markers are used and barrels are often struck by vehicles. The system is designed so that each individual robot is simple, reliable, and inexpensive. The proposed approach may have a higher equipment cost than traditional systems, however, there are possible cost reductions in labor and increased worker safety. The approach is not fitted for every work zone but there are many situations where it would be useful and practical. BACKGROUND Accidents can occur because of improper work zone design, improper work zone housekeeping, and driver negligence. Steps are being taken to improve work zone design. Many agencies now require written safety plans. Also, technology is being used to improve communication in work zones, leading to better work control [1]. Work zone housekeeping may be the most important element in reducing accidents [1]. A good safety plan is not helpful if it is not properly executed. Work zone housekeeping involves tasks such as covering and uncovering signs and moving markers (barrels, cones, signs) as the work progresses. While housekeeping is important, it comes at a very high cost. Worker time is wasted setting up and tearing down hundreds of construction zone safety devices. Safety markers often need to be deployed and retrieved at the beginning and end of each workday. Removing the markers or covering them is time-consuming, but leaving them in open view around the clock adds to unsafe conditions. Safety markers that are simply moved to the side of the road remain highly visible. The There are many ways to increase worker safety. For example, every road under repair could be completely closed to public traffic. Another solution may be to simply increase the number of safety cones and barrels in work zones to better channel traffic. While these solutions decrease the threat of injury, the cost of materials, workers’ time, and driver inconvenience is large. Obviously a reasonable balance must be found between worker safety, cost, and driver convenience. RELATED RESEARCH There is much research investigating “smart” highway systems and work zones. A large body of research exists on the Intelligent Vehicle Highway Systems (IVHS). This system uses information technology to provide information to travelers, improve traffic control and congestion, and increase the efficiency of commercial vehicle and transit operations [3]. Current IVHS research does not address traffic control through construction and maintenance areas. Although there is great potential for integration of IHVS and smart work zones. Several projects have applied robotic technologies to highway construction equipment. One project created a Remotely Driven Vehicle (RDV). The RDV is a radiocontrolled dump truck that is used as a shadow vehicle during slow-moving maintenance operations [5]. The shadow vehicle follows the work crew to serve as a barrier. The RDV removes the driver from being exposed to rear-end collisions while driving the shadow vehicles. Another highway construction robotics research project is the development of automated paving machines [6]. This system has increased the speed of these machines as well as the quality, and the smoothness of the pavement. The Advanced Highway Maintenance and Construction Technology (AHMCT) Research Center at the University of California, Davis has developed several automated highway maintenance systems. One system is an automated laser-guided lane striping machine [7]. This system is a traditional lanestripe-painting vehicle that uses sensors to partially automate the stripe painting process. A second project creates tethered robots for roadway crack sealing [8]. Normal crack sealing is a dangerous, tedious, and slow process. In this system, a tethered robot follows a work vehicle to rout pavement cracks prior to 2 Copyright © 2002 by ASME sealing. Other vehicle mounted robots can then seal the cracks. A third system is a flying aerial bridge-inspection system with a remote operator. This system inspects difficult to access areas on bridges using a video camera [9]. Another research group has developed a robotic paintremoval system for bridges. This system consists of a robot placed at the end of a long boom. The robot removes paint while collecting debris. This prevents the removed lead paint from polluting the environment [9]. There has been research in the automated placement of safety devices. A tele-robotic system has been developed to apply raised, reflective pavement markers [10]. Previously, the markers were applied by hand, with the worker positioned in a low-riding bay on a truck. This process was very time consuming and very dangerous since the markers were applied by the worker extending their hands into oncoming traffic. Two other systems automate the placement of safety devices with a safety cone dispenser. These systems automatically and quickly deploy and remove rows of warning cones [11, 12]. The systems are modified maintenance vehicles (pickups) that place cones as the vehicle travels down the road. The vehicle can then drive along the line of cones and retrieve the markers. Cones can be placed on either side of the vehicle and at various distances between cones. These projects have focused on smaller warning cones. Larger cones and barrels are often used so the work zone is visible from a grater distance and from the inside of a passing vehicle. These pervious systems function in a fundamentally different way than the proposed approach. The previous systems place markers in a serial fashion as the deployment vehicle is driven. The advantage of this approach is that the markers themselves do not need to be modified. One "centralized" robotic system is used to deploy/retrieve the markers. The cost and complexity of the overall system is not related to the number of markers deployed. The disadvantage of these systems is that the markers are deployed serially so the time required for deployment is linearly related to the number of markers placed. Also, the location of the markers can not be changed once they are deployed. Therefore the system cannot adapt to changes in the work zone. The proposed approach decentralizes the robotic system so the markers themselves become "smart". The advantage of this approach is that the markers can adapt to changes in the work zone. If a crew moves, the safety markers can follow and always maintain an optimal location. The markers can also be moved/deployed/retrieved in parallel decreasing deployment time. The disadvantage of this approach is that the cost of the overall system linearly scales with the number of markers used. Both approaches have advantages and there are work zone situations where each would increase worker safety and efficiency. The centralized approach (a single robot that deploys numerous barrels) would lend itself to situations where many barrels that are seldom re-configured are used. The decentralized approach (where each barrel is a robot) would be useful in situations where fewer barrels are used and are often moved. For example, a situation might require a wedge of barrels to follow a crew that patches a pothole every quarter mile. Individual mobile barrel robots could direct traffic quickly and efficiently without endangering workers and wasting worker effort. The robots could self-deploy, self-retrieve, and self-reconfigure. SYSTEM DESIGN In this project three prototype robotic safety barrels were designed, built and are being tested. The objective is to demonstrate the usefulness in increasing worker safety and decreasing traffic congestion delays through a more efficient, better-marked work zone Design Objectives There are several physical requirements that a robotic safety barrel must meet. The robot needs to be at least as stable as a traditional safety barrel. The robot and barrel need to remain stationary in 55mph winds and weigh less than 50 lbs. (so they can be moved by workers). The prototypes must have the same footprint as traditional barrels so current transport methods can be used. The device needs to cause minimal damage to vehicles if an impact occurs. It is important that the robots do not become a safety hazard. The robot needs to climb slopes up to 7% grade and travel at 5mph (the speed of some maintenance operations such as striping). The robot needs to traverse obstructions, bumps, and depressions. The robot needs to operate in inclement weather. The robot also needs to maneuver within the work zone. There are two systems-level design requirements that need to be addressed if the robot barrels are to be successful: 1) high reliability and 2) low cost. The system must be very reliable in hardware and software. A robot that malfunctions could enter into traffic and create a significant hazard. The system should be designed so critical components have multiple redundancy. The multi-robot aspect of the approach will be utilized to increase reliability-all robots could monitor their neighbors. Also, an encoded Radio Frequency (RF) carrier wave will be broadcast in the local area so all robots could be stopped in an emergency. There is extensive work in fault tolerant engineering. Robots are now being used in critical operations such as surgery and many systems with human interaction. All reliability issues were not addressed with the prototype robots but are being fully considered in design of the next generation. 3 Copyright © 2002 by ASME Cost per robot is also critical in a decentralized approach to robot safety markers. Multiple safety markers are used in work zones so costly markers are not practical. Also, barrels are often destroyed when struck by passing vehicles. Safety markers with a high replacement cost are not practical. However, robots struck by cars have performed their function by indicating to the driver that the vehicle is not in the correct position. Mechanical and Mobility Design The robot mechanical design can be seen in Figure 2. The robot consists of a platform that is similar to the base of a traditional safety barrel. The robot has two 7 ½” in diameter wheels that are independently driven by two motors. The rear of the robot has a passive caster. This kinematic arrangement allows the robot to turn on any radius including turning in place. The motors each have a high gear ratio so the robot is not easily moved when the motors are not powered. This allows the robot to stay in place for long periods while using little or no power. Since both the lead wheels are powered the robot can climb small bumps and obstacles. The barrel snaps onto a ring around the robot and encloses all components. The robot itself stands less than one foot tall and raises the barrel height by three inches. robots or by a remote planning system. This allows the robot processors to be simple, inexpensive, and redundant. CONTROL OF THE ROBOTIC SAFETY BARRELS The robotic safety barrels can be operated in many ways, ranging from being simply remotely driven (tele-operation) to operating autonomously. One mode of control that has been implemented on the robot prototypes is remote driving. This simple approach is advantageous because workers can deploy safety markers from a safe remote location. A joystick is used to drive the robot to its desired location. The base transmitter sends out encoded commands that contain a robot identification number and two desired wheel speeds. Each robot is constantly listening for commands. The operator can change the robot identification number and drive all robots individually. Remote control is simple and increases worker safety by moving the operator away from danger. However, it still requires worker time and robots are deployed serially. An autonomous system would be preferred. Autonomous control is broken into two problems- global planning and local control. The problem is divided in this manner to reduce the per robot cost. With this approach each robot does not need to possess high computational ability. Also, this approach requires only low-bandwidth communication with the central controller (or other robots). Global Planning Motion planning for mobile robots has been extensively studied. Several approaches have been developed such as behavior control [13], electrostatic potential fields [14], probabilistic maps [15], genetic planning [16] and others. Figure 2: Robot Mechanical Design Electrical Design The robot is powered by a 12-volt lead acid battery (shown mounted above the caster in Figure 2). This application lends itself to solar recharging, however, this option was not considered because of the desire to reduce the per robot cost. The barrel robot has two processors-one is used for robot control and one to manage communication with the outside world. Each robot has an RF transceiver that can transmit and receive information. Each motor has an encoder that can be used to detect wheel position or speed. The robot is designed to perform local control (wheel speed and/or position) with highlevel planning being done either by the distributed system of Previous work with multiple mobile robot control for highway maintenance has been done for crack sealing robots [8, 17]. Here, up to four tethered mobile robots work together. This system uses a host computer as the central intelligence and each robot becomes a workstation in a Windows network. The robots can access and share files but each remains independent. This is done to create a highly fault-tolerant system. This approach cannot be directly applied to the safety barrel robots because each robot will not have high computational ability. Also, the robots will not be tethered making high bandwidth communication expensive. Other research develops a hybrid control scheme for formations of non-holonomic mobile robots [18]. This method controls the motion of many robots while maintaining a predetermined formation. This work also uses high bandwidth communication and complex vision sensors on each robot. The approach being developed for the robotic safety barrels uses a single camera mounted on a maintenance vehicle. This 4 Copyright © 2002 by ASME camera would provide a view similar to the simulation image shown Figure 3. The video system identifies the location of each barrel robot with respect to itself using the highway stripes as reference. The control system then transmits a desired position to each robot. The robot itself moves to that position using its local control system. For example in Figure 3, robot #4 must move 1 meter forward and 1 meter to the left to join the formation. The desired positions are created based on a planning algorithm that dictates a series of desired positions along a path. The path can be created in many ways, here a polynomial is used. The boundary conditions for this polynomial are the beginning location and the desired end points and the initial derivative (the robot begins with locally forward motion), see Figure 5. This approach ensures the non-holnomic constraints are considered. The final orientation of the path is not constrained so a new destination can be given while maintaining fluid motion. Figure 3: Control Simulation The global control scheme is currently under development in simulation. The local controller, discussed in the following section is currently being tested in a laboratory setting. Other research has studied the use of video to identify barrel shaped objects [19] and the use of cameras for robot control in outdoor environments [20]. Local Control of Individual Robots Figure 5: Path Planning The control scheme also requires a relationship between differential motion in Cartesian space and differential motion in joint space. This jacobian relationship is nonlinear for the robot safety barrels. To find this differential relationship, consider the top view of the robot in Each barrel robot will receive a waypoint (i.e. desired position or desired velocity) form the central controller. The robot will then use a local control to obtain that position. A simple inverse Jacobian Cartesian scheme that uses PID control has been implemented. This is a common approach to robot control where an error (δx) is created in Cartesian space using a desired Cartesian position (xd=[xd, yd, ψd]) and a measured position (x). This error is transformed into joint space (δθ) using the inverse jacobian. This error is then acted upon by a traditional PID control law to create left and right torque commands (τ=[τl, τr]) for the robot's motors. Control of this type has several well known limitations (ignores dynamic effects, non-holonomic constrains) but initial tests have shown acceptable results (5-10% position error, e.g. 5-10 cm in a 1 meter motion) for barrel placement. Figure 6. Here the robot has moved an amount δx=[δx, δy, δΨ] between times ti-1 and ti. During this time the wheels have rotated an amount δθ=[ δθl, δθr]. Figure 4: Inverse Jacobian Cartesian Controller 5 Copyright © 2002 by ASME position of the robot is determined by accumulating differential motions, equation ( 4 ). X t = t i = X t = t i −1 + δX (4) This process (dead-reckoning) accumulates measurement errors. Laboratory testing has shown this error to be within 510% of the overall motion. This error accumulation is not critical since the global motion of the robot is determined by the absolute external measurement of the video system used by the global planner. Figure 6 : Differential Robot Motion During this differential time the robot will turn about a given radius by an amount δΨ, see Figure 7. The control scheme has been implanted and the results are shown in Figure 8. The robot was asked to travel to a position [x=1 meter, y=1 meter] in three seconds. The desired path is shown as a solid line and the actual path is shown with a dashed line. The robot arrives at the desired position with 9.2% position error. Figure 7: Turning Radius Here, SL and SR are the paths of the left and right wheels. No slip is assumed so the path lengths are related to the change in wheel angles by S=δθr where r is the wheel radius and δθ is given by the motor encoders. D is the distance between the wheels and RL and RR are the turning radii for each wheel. If small motions are assumed the differential motion in Cartesian space can be found based on the differential motions in joint space. Using geometry, the relationships are given by: δx = δy = r 2 [δθ L + δθ R ]⋅ cos r (δθ R − δθ L ) d (1) r [δθ L + δθ R ]⋅ sin r (δθ R − δθ L ) 2 d r δ Ψ = (δθ R − δθ L ) d (2) (3) The position of the robot in Cartesian space can only be determined from the changes in wheel position for small motions of the robot. For large motions the Cartesian position of the robot is non-observable based on wheel angle. The Figure 8: Desired and Actual Robot Path FUTURE WORK Future work on the robotic safety barrels will focus on increasing the system's reliability, decreasing the per robot cost, and the development of a global control scheme for a group of robots. The three prototypes cost approximately $700 per robot. Cost was not explicitly considered in their design and off-theshelf components were used. Almost half the cost of the prototypes was spent on the two drive motors. It is estimated that the next generation prototypes will cost around half this amount. It is hoped that a mass production approach can bring the cost per robot to less than $200. The next generation robots will also directly consider fault tolerance with triple redundancy in every critical system. Each robot will have several watchdog timers that must be frequently 6 Copyright © 2002 by ASME reset or the system will be shut down. The central controller will broadcast a carrier wave so that all robots can be stopped if a fault is detected. The robots may include three independent processors that will vote on all commands. These and other approaches will be directly incorporated into the next generation design. Other control approaches are being investigated. Namely, a fully autonomous mobile robot is being prepared to serve as the centralized controller. This robot is much more capable (large computational ability, GPS, laser range finder, stereo camera, etc) and will be a leader of several barrel robots. This general/troops system has the potential for more autonomy. SUMMARY AND CONCLUSIONS This paper presents a mobile robotic safety marker system that uses mobile robots to transport safety markers for highway construction and maintenance. These robots work in teams to provide traffic control. Independent autonomous barrel robots have several advantages. First, the barrels can self-deploy-eliminating the dangerous task of manually placing barrels in busy traffic. The barrel positions can be quickly and remotely re-configured as the work zone changes. Barrels could continuously follow work crews to maintain optimal placement. This paper described the motivation for the robotic safety marker system and how it could improve work zone safety. The design of three robot prototypes was described. A control architecture was briefly introduced and a jacobian matrix derived. The Robotic Safety Barrels are the first elements of a team of Robotic Safety Markers (RSM) that includes signs, cones, and possibly barricades and arrestors. Such devices are not suitable for every work zone, however there are many situations where they would useful and practical. ACKNOWLEDGMENTS This work is sponsored by the National Research Council Transportation Research Board’s NCHRP IDEA program. REFERENCES [1] Anon. “The 11 Best Ways to Improve Work Zone Safety.” Better Roads v60, n7 (1990): p20-23. [2] Ha, T.J., and Nemeth, Z.A. “Detailed Study of Accident Experience in Construction and Maintenance Zones.” Transportation Research Record n1509 (1995): p38-45. [3] Ravani, B., and Velinsky, S.A. West, T.H. “Requirements for Application of Robotics and Automation in Highway Maintenance and Construction Tasks.” ASCE Specialty Conference on Robotics for Challenging Environments (1994): p356-364. [4] Gumtau, Richard. “Richard Gumtau on Traffic Control and Work Zones.” Better Roads v63, n9, 1993. [5] Shah, Shashikant. “New Work Zone Safety Devices.” Proceedings of the ASCE 3rd International Conference on Applications of Advanced Technologies in Transportation Engineering (1993): p308-315. [6] White, Thomas D. “Evolving Automation in the Asphalt Paving Industry.” TR News n176 (1995): p4-6. [7] West, Thomas H., and Velinsky, Steven A., and Ravani, Bahram. “Advanced Highway Maintenance and Construction Technology Applications.” TR News (1995): p17-23. [8] Lasky, Ty A., and Ravani, Bahram. “Sensor-based Path Planning and Motion Control for a Robotic System for Roadway Crack Sealing.” IEEE Transactions on Control Systems Technology v8, n4 (2000): p609-622. [9] Woo, Dah-Cheng. “Robotics in Highway Construction and Maintenance.” Public Roads v58, n3 (1995): p26-30. [10] Rihani, Rami A., and Bernold, Leonhard E. “Telerobotic Pavement Marker Application.” ASCE Conference on Robotics for Challenging Environments (1996): p171-177. [11] Zhou, Tong, and West, Thomas. “Assessment of the Stateof-the-Art of Robotics Applications in Highway Construction and Maintenance.” Proc. of the 2nd Intrl. Conf. on Applications of Advanced technologies in Transportation Engineering (1991): p56-60. [12] AHMCT website: http://www. ahmct.ucdavis.edu/ [13] Brooks, R., "A Robust Layered Control System for a mobile robot," IEEE Transactions on Robotics and Automation, Vol. 2, No. 1, 1986. [14] Khatib, O. "Real-time obstacle avoidance for manipulators and mobile robots," International Journal of Robotics Research, 5 (1), 90-98, 1986. [15] Borenstein, j., Koren, Y., "The Vector Field HistogramFast Obstacle Avoidance for Mobile Robots," IEEE Trans on Robotics and Automation, Vol. 7, No. 3 June 1991. [16] Farritor S., Hacot H. and Dubowsky S., "Physics-Based Planning for Planetary Exploration" IEEE Intl. Conference on Robotic and Automation, 1998. [17] Feng, Xin, and Velinsky, Steven A. “Development of a Distributed Multiple Mobile Robot Control System for Automatic Highway Maintenance and Construction.” Midwest Symposium on Circuits and Systems, Proceedings of the 1997 40th Midwest Symposium on Circuits and Systems v1 (1997). [18] Fierro R., Das A.K., Kumar V., Ostrowski J.P., “Hybrid Control of Formations of Robots,” IEEE Intl. Conference on Robotic and Automation, 2001. [19] Hood, F. Hoff, W., King, R., "Evaluation of an Interactive Technique for Creating Site Models from Range Data", ANS Meeting on Robotics and Remote Systems, 1997. [20] Ollis, M., Stentz, A., "Vision-Based Perception for an Autonomous Harvester" Proceedings of the IEEE/RSJ International Conference on Intelligent Robotic Systems, Vol. 3, 1997, pp. 1838 - 1844. 7 Copyright © 2002 by ASME