Chain-of-Causation

Chain-of-Causation is a way for a self-driving car's AI to explain why it did something by linking cause and effect together, one step at a time. It's similar to how a learner driver, when asked "why did you brake?", might answer in sequence: "the car ahead suddenly stopped → that created a risk of collision → so I braked."

A self-driving car decides where to go next (its path) based on what its cameras see, but instead of just outputting that decision alone, it also records the situational understanding and logic that led to it. This causal record can later be used by humans to check why a particular driving decision was made, or reused as training data.

The article explains that NVIDIA's autonomous driving model "outputs future trajectories together with chain-of-causation reasoning traces." It's worth noting this isn't a human-friendly driving diary, but rather data in which the model structures and records its internal decision-making process.