Adapting US Army Acquisition to Ensure the Reliability and Safety of Autonomous Vehicles

This report presents several challenges that the U.S. Army will face in the transition to autonomous vehicles, challenges that are only magnified in the current acquisition environment with limited testing. Artificial intelligence algorithms introduce additional complexity, resulting in systems with a complex combination of human, machine, and autonomous controllers.

Figure 1. Sample control structure of an autonomous system.

Artificial intelligence (AI) has become prevalent in many fields in the modern world, ranging from vacuum cleaners to lawn mowers and commercial automobiles. These capabilities are continuing to evolve and become a part of more products and systems every day, with numerous potential benefits to humans. AI is of particular interest in autonomous vehicles (AVs), where the benefits include reduced cognitive workload, increased efficiency, and improved safety for human operators. Numerous investments from academia and industry have been made recently with the intent of improving the enabling technologies for AVs. Google and Tesla are two of the more well-known examples in industry, with Google developing a self-driving car and Tesla providing its Full Self-Driving (FSD) autopilot system. Ford and BMW are also working on their own AVs.

These systems continue to remain a challenge despite these investments. Recent National Highway Traffic Safety Administration (NHTSA) investigations involving Tesla highlight the importance and difficulty of these technologies. NHTSA Campaign Number 23V085000 (NHSTA, 2023a) involves the recall of over 300,000 vehicles equipped with the FSD Beta system. The recall summary states that the FSD Beta system may result in unsafe actions with the vehicle. These unsafe actions include traveling straight through an intersection while in a turn-only lane, entering a STOP sign-controlled intersection without coming to a complete stop, and proceeding without caution into an intersection during a steady yellow traffic signal. The recall also states that the system may fail to respond to changes in posted speed limits and not account for the driver’s adjustment of the vehicle’s speed beyond these limits.

NHTSA Campaign Number 23V037000 (NHSTA, 2023b) was an earlier recall involving approximately 53,000 vehicles that also discusses issues with the FSB Beta system. The system included a “rolling stop” functionality in the software that could allow the vehicle to travel through an all-way stop intersection without stopping. Both recalls were able to be addressed through remote Over-the-Air software updates, which is actually an important benefit that AVs will likely have. A recent survey of deep learning methods for AVs also recognized the difficulties that remain for widespread use of AVs. The work identified 10 challenges that must be resolved: 1) Complexity of Autonomous Driving (AD) Systems; 2) Dynamicity of Road Environment; 3) Big Data and Real-Time Processing; 4) Intelligent Data Prioritization; 5) Robustness and Adaptability; 6) Integration/Fusion of Sensory Data for Dynamic Decision Making; 7) Fairness, Accountability, and Transparency in deep learning for AD; 8) Online Learning Capabilities in AD; 9) Robustness Against Adversarial Attacks; and 10) Variability of Traffic Sign Boards (Muhammad et al., 2021).

According to NHTSA guidance, the overall safety of AVs is left to the companies that build them. Each company must comply with Federal Motor Vehicle Safety Standards while also certifying that their vehicles do not have inherent safety risks. While many companies are currently testing vehicles with higher levels of automation to ensure that they operate as intended, the NHSTA indicates that numerous experts state that more work remains to be done by vehicle developers to ensure their safe operation before they are available commercially (NHSTA, n.d.). Detailed studies on the algorithmic methods for safe autonomous driving have not yet been completed. These studies should be considered the backbone of the safety of AVs, including those developed for use by the Army.

This work was performed by Patiana Theragene, Martin Wayne, and Nathan Herbert for the U.S. Army DEVCOM Analysis Center. For more information, download the Technical Support Package (free white paper) below.

This Brief includes a Technical Support Package (TSP).
Document cover
Adapting US Army Acquisition to Ensure the Reliability and Safety of Autonomous Vehicles

(reference DEVCOM-2023048) is currently available for download from the TSP library.

Don't have an account? Sign up here.