Boeing

Low-Cost Variable-Autonomy Ground Vehicles

DARPA Urban Challenge vehicle

This DARPA Urban Challenge vehicle had an overhead electric generator to power various control systems, in order to provide the needed electrical power. Future unmanned ground vehicles must address questions such as how to generate enough power through a compact source.

Industry has made substantial advances on unmanned aerial vehicles. But what about unmanned ground vehicles?

The year 2016 has been a particularly productive year for the commercial implementation of autonomous ground vehicle technology. Various automakers are racing to introduce autonomous transportation to the consumer market.

The U.S. National Highway Traffic Safety Administration released a long-awaited policy in September 2016 to lay a regulatory foundation for widespread use of autonomous vehicles on public roads. Together, both technological and regulatory needs must be solved for practical widespread use for autonomous ground vehicles.

While autonomous vehicle technology will eventually be safe and reliable, this is likely not the case in the near future. It remains to be seen how the rushed deployment of autonomous vehicles on public roads plays out, with all the risks associated with the technology.

However, within the constraints imposed by the state of the art, particularly in regards to reliability, there are military applications that are feasible today. Unmanned Ground Vehicles (UGVs) have made considerable strides in the past decade since the three DARPA Grand Challenge competitions that were held in 2004, 2005 and 2007. The last of these was the DARPA Urban Challenge and specifically focused on the ability of autonomous ground vehicles to interact with other unmanned and manned vehicles.

While DARPA’s primary objectives were achieved, some of the limitations of present-day technology were also highlighted. One of the more significant of these limitations is best illustrated by the autonomous vehicle from the Massachusetts Institute of Technology. The MIT team was one of the better performers in the contest, having successfully completed the designated course in the finals of the competition.

As can be seen in the photo at the top of this page, though, the power requirements of this vehicle’s sensor and processing hardware was so extensive that it required an independent gasoline-powered generator strapped to the roof for the needed electrical power. With the interior of the vehicle also entirely filled with relevant electronics, the payload capacity of this vehicle is considerably limited. Moreover, the expensive control hardware on board the vehicle far exceeds the cost of the vehicle itself. This limitation is not confined to the MIT vehicle; some of the teams participating in the Urban Challenge had spent in excess of $10 million in developing their vehicles.

While this concept of “performance-at-any-price” may work in winning contests, the practical world demands much lower price points to enable adoption. One of the purposes of the DARPA Grand Challenge contests was to develop an autonomous supply vehicle for use in war zones. In this era of tight defense budgets, it would be difficult to market million-dollar autonomous vehicles as potential replacements to manned supply vehicles, even if all the capabilities are available.

The second major limitation highlighted at the DARPA Grand Challenges is in vehicle capability. It was seen repeatedly throughout the 2007 Urban Challenge that despite all the expensive hardware and elaborate software, vehicles were immobilized as a result of their inability to process a scenario that any human driver would have considered trivial. One such example was when a gate that was supposed to have been kept open unexpectedly closed, blocking a lane of the roadway. The MIT vehicle came to a complete stop when it encountered the gate blocking its lane, but would not change lanes and pass around the gate. It was only upon human intervention in the form of a crew that arrived at the scene and opened the gate that the MIT vehicle was able to continue on its mission.

While most of the participants at the DARPA Urban Challenge 2007 comprised of vehicles with very expensive control hardware and software, there were some exceptions. Most notably, a team from Kokomo, Indiana, developed an autonomous vehicle for under $20,000. The key enabler was a simplified vision and GPS sensor system that made use of commercially available off-the-shelf, lower-cost hardware. At the bottom of this page is a photograph of the autonomous vehicle developed by this team (a semi-finalist), and as can be seen, many of the vision sensors were $50 webcams intended for use with personal computers.

However, this system can, on some occasions, produce false positive results, i.e., the system detects an object that it classifies as an obstacle whereas in reality it is not of any concern. It should also be noted that the vision system, while providing a practical and effective input to the vehicle control system, is not designed to work in all lighting conditions. Performance can be degraded under very low or very bright ambient light. Under such conditions, the control system would have to shift to using alternate sensor input, such as GPS and short-range radar.

Despite the unique innovations, all vehicles were subjected to the same limitations— occasional inability to operate in seemingly trivial traffic situations.

It is not sufficient from the practical perspective that autonomous vehicles are able to handle 99 percent of the scenarios and fail whenever the remaining 1 percent is encountered. As seen in the 2007 DARPA Urban Challenge and cited earlier, million-dollar autonomous vehicles were reduced to a state of immobility when encountering common traffic scenarios that would have been easily overcome by a human driver.

The most practical means of overcoming this is represented in a patent issued to Boeing in 2015. That is, couple autonomous ground vehicles with a remote human operator who is able to remotely intervene whenever the vehicle encounters a situation that it is not able to overcome. For optimal performance, the level of involvement of the human operator is variable and entirely situation-dependent. Moreover, the transition of control between the autonomous vehicle control system and the human operator, and vice-versa, should ideally be seamless. This type of hybrid vehicle control utilizes the best of both realms and leads to a new class of vehicle – the Variable- Autonomy Ground Vehicle (VAGV).

While it may superficially appear that the VAGV concept is a step backwards in the development of autonomous vehicles, this concept is not entirely new, as can be seen from the Predator UAV control system. Although it has a limited level of autonomous operation, the remote human operator makes all the important decisions, particularly in regard to discharge of weapons. A similar control model is being extended here for autonomous ground vehicles, where remote operation through a high-bandwidth radio channel permits the human operator to manually take over control on a temporary basis until the obstacle or situation encountered is overcome.

The remote operator can be physically located at any convenient location which permits communication and control of the VAGV over a secure radio channel. This can be anywhere from the immediate proximity of the vehicle to the other side of the world, as is the case with the Predator UAV. Because the VAGV is able to maintain autonomous control most of the time, it is possible for one operator to handle multiple VAGVs in the field. The primary function of the operator is to intervene when requested by the VAGV navigation control system and to take manual control, if needed, of the VAGV’s surveillance, fire control or other additional systems, when it is so equipped.

The main mode of communication from the VAGV to the operator is in the form of a live video feed from various onboard cameras. These can also include infrared imaging cameras for night use.

While not essential for most missions, a Virtual-Reality helmet used by the remote operator, in which all the visual information streaming from the VAGV is appropriately projected inside the helmet, would provide a level of situational awareness that is second to only being present in the vehicle.

Using the combination of a proven low-cost approach, coupled with a variable level of autonomous control, appears to be the most practical means of delivering reliable unmanned ground vehicles capable of filling a variety of roles. Among the possible applications are autonomous supply vehicles in war zones, as casualty extraction vehicles in certain circumstances, as ISR platforms with long range and endurance capability, as border patrol surveillance tools and even as assault vehicles capable of carrying a variety of armaments. Moreover, with the additional functionality of interoperability with other unmanned and manned air and ground systems, a significant enhancement in overall capability is achieved.

A rapid expansion into the adjacent area of unmanned/autonomous ground vehicles, particularly for the applications described here, is well within current technological reach and resources. 

To read and download the complete Boeing Technical Journal paper titled “Low-Cost Variable-Autonomy Ground Vehicles,” click here.

By Mahesh K. Chengalva