2.2 State-of-the-art in autonomous systems
2.2.3 Defence and homeland security
As a key driver for the development of autonomous systems, defence and homeland security needs have led to the deployment of many unmanned systems, especially aerial and ground vehicles intended for combat, reconnaissance, explosives ordnance disposal, as well as search and rescue.
Future Combat Systems and the Joint Robotics Programme
The U.S. Department of Defence (U.S. DoD) envisions a strategic transition to a
“full-spectrum” force capable of being deployed at short notice to any location in the world (Rose et al., 2002). In line with this vision, vehicles with robotic capabilities are expected to form a third of all ground forces by 2015 (Bertozzi et al., 2006). This is further driven by heightened global security concerns due to the threats of terrorism.
To achieve this vision, efforts are directed at the development of “next-generation”
unmanned vehicles with enhanced deployment ease and reconfiguration flexibility for situations of conflict without endangering human lives (Lopez, 2005). These initiatives fall under the purview of the Future Combat Systems (FCS) programme, and the Joint Robotics Programme (JRP). FCS is formed by a force of lighter, more responsive units that are both lethal and survivable (Rose et al., 2002). It comprises different mobile platforms with autonomous capabilities that can be readily deployed, ranging from pint-sized, man-transportable ground and aerial robots (Bruch et al., 2005), Multifunction Utility/Logistics Equipment (MULE) vehicles, and large-scaled unmanned ground vehicles, such as the Armed Robotic Vehicles (ARVs), the latter which are being planned for deployment at a more mature stage (Lopez, 2005). An example is the Demo III XUV testbed, as shown in Figure 2.1(c), which demonstrates the current level of autonomy realizable in off-road conditions. In comparison, the need for smaller
vehicles for surveillance, reconnaissance, or explosive ordnance disposal (EOD), are being met by commercial robots such as the Packbot by iRobot, as shown in Figure 1.1(a).
The challenges faced by current systems developed for FCS differ from those of truly autonomous systems; many unmanned vehicles are designed with remote teleoperation as the primary mode of operation, with limited or partial autonomy while keeping human operators within the control loop (Rose et al., 2002). However, the absence of an onboard operator demands heavier emphasis on prognostic and diagnostic information to be provided by unmanned vehicles to the operations centre (Lopez, 2005). Apart from this, the perception systems used, and the resulting system architectures share many similarities with those designed with autonomous operation in mind. Therefore, a good understanding of these systems is still relevant for the design of autonomous systems, for the purpose of which these tele-operated systems are an intermediate milestone.
The DARPA Grand Challenge
As the strategic research arm of the U.S. Department of Defence, the Defence Advanced Research Projects Agency (DARPA) funds several competitions to spur the development of enabling technologies for autonomous systems. Among these, the DARPA Grand Challenge (DGC) has been the most successful. DGC is an outdoor competition in which fully self-driven, unmanned ground vehicles attempt cross-country navigation across off-road, unrehearsed terrain within a time limit. The first grand challenge in March 2004 exemplified the state-of-the-art in autonomous vehicle technology currently achievable. However, the harsh environmental conditions were left largely unconquered;
fifteen vehicles attempted the race and none finished. Many vehicles came to a halt as a result of slopes, negative obstacles, undetected foliage, errors in perception or planning, as shown in Figure 2.2. The frontrunner, “Sandstorm” from Carnegie Mellon University (Figure 2.1(d)) completed 7.4 miles, before going off-course.
In comparison, the 2005 grand challenge signified the substantial progress made in the development of autonomous ground vehicles capable of travelling long distances across difficult, unstructured and obstacle-laden terrain (Seetharaman et al., 2006). The competitors were clearly more successful than those of 2004; only one vehicle failed to pass the 11.84 km (7.4 mile) mark set by “Sandstorm” in the previous year. The winning vehicle, “Stanley,” from Stanford University, completed the course under seven hours with the aid of multiple laser range finders and proximity sensors that assisted the vehicle in building traversability maps of the immediate terrain (Thrun et al., 2006).
(a) (b)
(c) (d)
Figure 2.1: Unmanned ground vehicles. (a) Artist’s impression of one of the Mars Rovers (Spirit and Opportunity). (b) INRIA’s CyCab testbed (Pradalier et al., 2005) (c) The DEMO III XUV Platform (Albus et al., 2002) (d) Sandstorm, by Team Red, one of CMU’s entries in the DARPA Grand Challenge 2004.
As in the case of “Stanley” and “Sandstorm”, prevalent approaches include drive- ability analysis of the terrain and formation of traversability maps in real-time, where such traversability models are supported by data obtained from different sensors (Thrun et al., 2006; Whittaker, 2005), e.g. colour images (Chaturvedi et al., 2001), range data from stereovision (Singh and Digney, 1999), or a combination of both (Quek et al., 2005). The use of multiple sensors is deemed necessary, as no single sensor is capable of observing the entire perceptual spectrum. The integration of different types of sensors would enable the vehicle to observe a larger spectrum of traversability (Touchton et al., 2006). With GPS coordinates, the problem becomes one of decision making based on the minimization of some cost function (e.g. the safety of the vehicle), given the traversability of the terrain (as obtained from sensors), and planning trajectories to reach the destination, with the aid of GPS waypoints.
The next milestone in the autonomous vehicle roadmap is to achieve operation within urban environments. Driving within urban environments is fraught with a different
(a) (b)
(c) (d)
Figure 2.2: Snapshots from the 2004 DARPA Grand Challenge. These images show how slopes, negative obstacles or undetected foliage and errors in perception or planning result in vehicles coming to a halt.
set of challenges; it requires more than precise, reliable, repetitive or robotic behaviour, and further entails selective attention, alertness and instinctive responses to varying road conditions, obstacles and safety conditions (Seetharaman et al., 2006). Vehicles are expected to interact with other road users while obeying rules and negotiating traffic, where the reaction time is decreased significantly as compared to a sparse dessert environment (Seetharaman et al., 2006). Apart from avoiding both static and mobile obstacles, navigation on-road requires perception of road signs and lane markings, and interaction with other road users (Srini, 2006). This demands autonomous vehicles with capabilities beyond those demonstrated in the past two grand challenges. To spur the development of vehicles with such capabilities, an urban version of the grand challenge (i.e. the DARPA Urban Challenge) was held in November 2007. There is a convergence between the objectives of the urban challenge, and those of the automotive industries which are leveraging on the perception and sensing technologies developed for autonomous systems to provide driving assistance and enhance the safety and efficiency of the transportation system (Bertozzi et al., 2006).
Robots for Urban Search and Rescue
It has been claimed that Urban search and rescue (USAR) presents a greater challenge than the DARPA Grand Challenges, given the extreme operating conditions and harsh environments (Murphy, 2004; Willmott, 2005). While most devices developed for USAR purposes are remote-operated, a conglomerate of robotic technologies is being sought to address the challenges of autonomous operation, focusing particularly on providing mobility within haphazard environments, perception, detection of human presence, and decision making. An instance of this integrated approach drives the research at the Centre for Robot Assisted Search and Rescue (CRASAR) in the University of South Florida, U.S.A., whose efforts are applied as part of a crisis response centre. In addition, the development of urban search and rescue robots is further supported by government agencies e.g. the National Institute of Science and Technology (NIST) in the United States, which has been establishing metrics and validation tests to assess the progress of these robots in a measurable and repeatable manner (Wang et al., 2003).
While much efforts are underway, none of the algorithms demonstrated by CRASAR or other organizations at various robotics search and rescue competitions are useable in a real disaster situation (Murphy, 2004). Rescue robots have to operate in environments which are haphazard and irregular, and where a priori information would be nearly- impossible to obtain. In addition to technical challenges in the operation of such systems within hostile environments, other practical issues need to be considered (Carlson and Murphy, 2005). These include acceptability issues and alignment with on-going rescue efforts (Willmott, 2005). To gain acceptance, robots being developed for urban search and rescue have to exhibit greater intelligence and autonomy in performing their tasks, or remain tethered to their human operators. They have to demonstrate innate survival capabilities, especially when any breakdowns may hamper the whole rescue operation.