Understanding the Challenges of Self-Driving Cars: Lessons from Military Drone Operations

Self-driving cars are touted as the future of transportation, promising safer roads and less traffic congestion. However, these autonomous vehicles often falter in everyday situations that human drivers navigate with ease. Scenarios involving construction zones, school buses, power outages, and unpredictable pedestrians frequently overwhelm their systems, leading to erratic behavior, crashes, or stalling. Notably, these failures can cause significant disruptions, blocking traffic and hindering first responders.

To manage these risks, many self-driving companies employ human “babysitters” who monitor the vehicles remotely, stepping in when the technology falters. This raises the question: have these companies learned from the military’s lengthy experience with unmanned aerial vehicles (UAVs)?

Remote Supervision: A Familiar Concept

The notion of humans supervising autonomous vehicles from a distance is not new. For decades, the U.S. military has operated UAVs remotely, learning valuable lessons about control systems, human-machine interactions, and the complexities involved in remote supervision.

As a Navy fighter pilot in the 1990s, I was at the forefront of researching ways to enhance these remote operation systems. My experiences, along with those of many others, highlighted crucial insights that are surprisingly relevant to the ongoing challenges posed by self-driving cars.

The Nature of Supervisory Control

Supervisory control in this context refers to humans overseeing and supporting autonomous systems. There are two primary forms of this control for self-driving cars:

1. Teleoperation: Here, a human remotely controls the vehicle, managing its speed and steering in real time. This method is highly sensitive to communication delays, which can result in dangerous lag times during critical maneuvers.

2. Remote Assistance: In this variation, instead of driving the car directly, a human provides guidance by laying out a path or interpreting complex situations that the AI struggles to understand. This method can tolerate more delay than teleoperation but is still time-sensitive.

Key Challenges: Lessons From Military Operations

The military has faced various challenges in UAV operations over the past 35 years, and these experiences provide essential lessons for the self-driving industry.

1. Latency

Latency, or delays in communication, is one of the most significant hurdles in remote vehicle control. Early drone operators in the military faced communication lags exceeding two seconds, resulting in an accident rate 16 times higher than that of traditional fighter jets. This delay made real-time control incredibly difficult and has parallels in the self-driving industry where reliance on cellphone networks can result in similar issues.

2. Workstation Design

The design of control interfaces has been critical in both military UAV operations and self-driving vehicles. Confusing controls or poorly designed displays have historically contributed to numerous incidents, with some military drone accidents attributed to interface problems. The self-driving industry mirrors this challenge, often using off-the-shelf gaming controllers for remote operation, which can lead to mode confusion during critical situations.

3. Operator Workload

Remote operators often experience extreme fluctuations in workload, oscillating between intense focus and boredom. This inconsistency can lead to errors, particularly when immediate attention is required during emergencies. The military’s attempt to have one operator oversee multiple drones highlights this issue, as cognitive switching costs can increase stress and decrease performance. Self-driving vehicles may encounter similar issues, requiring more robust modeling of operator capabilities and workload management.

4. Training

Training protocols in early military drone programs were often inadequate, resulting in accidents. It wasn’t until later that comprehensive analyses of necessary knowledge and skills led to improved training programs. The self-driving industry lacks transparency about operator training standards, raising concerns about the effectiveness of remote operators who play critical roles in ensuring on-road safety.

5. Contingency Planning

Successful aviation operations have well-defined protocols for emergencies, including strategies for lost communication and potential failures of autonomy. The self-driving sector appears less prepared for such contingencies, as evidenced during power outages when self-driving cars failed to execute safe maneuvers, blocking traffic and creating hazardous situations.

Moving Forward

The history of military drone operations is rich with lessons that the self-driving car industry can apply. The key takeaways—low latency, well-designed interfaces, manageable workloads, rigorous training, and thorough contingency planning—are essential for safe remote supervision. Unfortunately, many self-driving companies seem to be repeating the early mistakes made by the military.

While autonomous technology continues to evolve, the reliance on human supervision will remain crucial as AI struggles to address uncertainties. The opportunity is ripe for the self-driving industry to learn from military lessons, ensuring safer roads for all.