Waymo has introduced a new cognitive system called ReD (Reference Driver), designed to enhance the safety of its robotaxis by simulating human driving behavior. This initiative aims to address the critical question of what makes human drivers more adept at avoiding accidents compared to autonomous systems.
Understanding ReD's Capabilities
The ReD model is based on the neuroscientific concept of active inference, which suggests that humans constantly strive to minimize surprises. This model simulates how a careful and competent human driver updates their beliefs as situations evolve, manages uncertainty about other road users' intentions, and selects evasive maneuvers such as braking or swerving. According to Waymo, this approach helps in devising a scientifically grounded method for evaluating collision-avoidance behavior.
Collaborative Development and Open Source
Waymo developed ReD in collaboration with the Delft University of Technology in the Netherlands. The findings were published in a Nature research paper, emphasizing the model's role as a behavioral crash test dummy designed to prevent crashes. The company is also working with safety organizations and regulators to refine the model further. To accelerate its adoption, Waymo plans to make ReD open source under an academic (non-commercial) license.
Key Features of the ReD Model
- Looming: Judges threats based on how fast an object grows in its field of view.
- Traffic Norm: Filters actions that fall outside of law-abiding behavior.
- Single-foot Driving: Accounts for a 0.2-second pause between gas and brake application.
The model also incorporates a proactive avoidance strategy, assuming that something will go wrong and anticipating potential risks to avoid conflicts.
Implications for Autonomous Vehicle Safety
The development of ReD represents a significant step forward in the quest for safer autonomous vehicles. By simulating human-like driving behavior, Waymo aims to improve the safety and reliability of its robotaxis. This initiative could set a new standard for evaluating and enhancing collision-avoidance systems in the autonomous vehicle industry.
"Evaluating AV safety is multifaceted, and understanding how a human handles conflict is a critical piece of the puzzle," said Waymo safety chief Mauricio Pena.
The collaboration with academic and regulatory bodies underscores the importance of a shared, scientifically grounded approach to autonomous vehicle safety. As ReD becomes open source, it could foster further innovation and collaboration across the industry.
Sources:
