Intelligent Vehicle Source,
April 1999

Demo III Autonomous Military Vehicle Beginning to Take Shape

These vehicles are primarily designed to handle off-road terrain. According to a program overview provided by RST, “the primary goal of Demo III is to develop and demonstrate technology required to develop a small, survivable mobile robot capable of autonomous operation over rugged terrain as part of a mixed military force, containing both unmanned and manned vehicles.” The overall motivation is to reduce casualties on the battlefield, particularly in scouting misions. In fact, the vision of the US Army, as defined in their “Army After Next” strategic plan, is for a large segment of all battle theater vehicles to be capable of autonomous operation. And, with recent experience with missions in urban areas such as Bosnia, an essential related requirement is the ability of the vehicles to traverse roads of all types, to include maneuvering amongst normal roadway vehicles. They are also required to convoy with one another on the highway, in a vehicle-follower fashion. Four vehicles will be produced for Demo III, all capable of interacting intelligently and cooperatively.

Latest in a Decade of Demos

The US Army Demo III program follows about ten years of earlier work sponsored by the Defense Advanced Research Projects Agency (DARPA), incarnated as Demo I and Demo II. As Demo II was proceeding in the mid-nineties, the technology began to mingle with highway applications, as organizations such as Carnegie Mellon University (CMU) and (then) Martin Marietta were active in both the DoD work and the DOT research in automated highway systems. As an example, CMU's vision-based road following algorithm, “RALPH” began its life as a DoD application -- it was further developed by the National Automated Highway System Consortium (NAHSC), and eventually put to use in the “Free Agent” scenario at Demo '97, in which Houston Metro buses and cars interacted under full automated control. Now, these algorithms are part of the standard robotic vehicle toolkit, and vision-based capabilities in Demo III build upon the NAHSC work.

Full Sensor Suite Helps Meet Rigorous Requirements

The Demo III Experimental Unmanned Vehicle (“XUV,” in program lingo) employs a wide variety of sensors for mobility. These include day/night stereo vision, a scanning laser rangefinder, and two radar systems, a 4 GHz radar to penetrate vegetation, and a 77 GHz radar for imaging obstacles at longer ranges. Maneuvering and path planning decisions will rely upon both multi-spectral and polarizing imagers to help differentiate between classes of obstacles such as vegetation, pavement, soil, rocks, etc. Additional small ultrasonic sensors will also be employed for close-in obstacle avoidance.

The challenges for off-road operation are many. The requirement to operate at 20 mph (32 kph) is actually fairly fast, even for even human drivers. The “brain” of the system must be agile and astute -- simple tasks for a human, such as detecting a narrow ditch or fencing, require a complex interaction of sensor pointing (targeting), sensor fusion, and software interpretation to make the best mobility decisions.

The Demo III vehicle must perform a variety of military scout functions, with autonomous mobility being simply a supporting function. These mobility requirements (see Sidebar) call for the vehicle to maneuver autonomously while avoiding all non-negotiable obstacles:

at speeds of up to 40 mph (64 kph) on primary roads during daylight hours and dry conditions;
at speeds up to 20 mph (32 kph) on primary roads during nighttime or wet weather conditions;
at speeds up to 20 mph (32 kph) cross country and on secondary roads and trails during daylight hours and dry conditions; and
at speeds of up to 10 mph (16 kph) cross-country and on secondary roads and trails during nighttime or wet weather conditions.
The vehicle is required to deal appropriately with a number of situations encountered, including automatically modifying vehicle speed to levels appropriate to terrain and vehicle dynamics; traversing slopes with grades of 60% fore and aft and 40% side slope; negotiating a twelve inch vertical step in forward and reverse direction; and negotiating a twelve inch gap. The vehicle is also expected to be able to detect humans that are within a few hundred feet of the vehicle and take evasive action so as to make it more difficult to disable.
Vegetation Penetration

When traveling off-road, the sensors must be able to detect negative obstacles (e.g., holes, ditches, etc.) as well as positive. Since the vehicle must be able to traverse vegetated terrain, sensors must have the capability to penetrate vegetation to detect obstacles, a requirement unique to the military mission. Further, since traversing cross country often involves dust, the sensors must be able to penetrate heavy dust. The obstacle avoidance system must also detect water to help the vehicle identify and negotiate puddles, ponds, creeks and rivers. The Demo III team plans for most of these off-road capabilities to be seamlessly integrated into on-road obstacle avoidance since the same sensor package provides coverage in both modalities.

To provide for safety while backing up, sensors cover a 12 foot (3.7 m) rear sensing range for reversing at up to 5 mph (8 kph). When the vehicle is convoying on the highway, side detection capability will be active, with a 10 foot (3 m) side sensing range to allow lane changes of up to one lane width. All obstacle range accuracies are required to be within 5% and sensor response time to any situation less than 1/2 second.

Autonomy on the Fast Track

Demo III, barely a year old, will have its first operational demonstration this September at Aberdeen Proving Ground in northeastern Maryland. The program focuses strongly on showing capability to the Army customers who must eventually decide whether to procure production systems and integrate them into tactical military operations. A second demonstration for this purpose will take place in September 2000, and the culminating event will be field testing by Army troops -- no engineers allowed -- in September of 2001. Capability will increase with each demonstration until all requirements are met. At a Demo III Interim Progress Review (IPR) in late February, the basic platform of the first vehicle was exhibited, which is being tested during March. Orders have been placed for sensors and subsystems, which will be integrated beginning in March. Software developers also planned to begin integration of software modules in March. The schedule calls for vehicle completion in July, and full shakedown testing in August.

21st Century Sticker Shock

Unofficial estimates for production versions of the XUV range around $500K per unit in small quantities, with the military mission sensors comprising the majority of that cost. The cost of the autonomous mobility capability then can be ballparked at around $200K, with costs dropping as volumes rise and the manufacturing design process evolves.

Tech Transfer Encouraged

Chuck Shoemaker of the Army Research Lab (ARL) in Adelphi, Maryland, is the Demo III program manager, with contract management provided by Jeff Jackowski of the Army's Tank and Automotive Command (TACOM) in Warren, Michigan. The Pentagon sponsors of Demo III are quite aware of the possibility for program spinoffs to commercial applications and are willing to share selected program results with researchers and product developers in the vehicle and transportation industries.

Key subcontractors to RST in this project are Science Applications International Corporation (SAIC) and Sarnoff Corporation (a subsidiary of SRI International). SAIC acts in a Deputy Program Management role and is focused on system integration, technology transfer, systems engineering, communications, field exercises, and software development for autonomous mobility and mission planning. Sarnoff specializes in the development of state-of-the-art image processing systems; their role in Demo III is technology transfer, vision algorithms for road following, motion detection algorithms, and mosaic compression routines for reconnaissance functions.

other articles...