Transit Design Runs to Thousands of Variants. Wu's Proposed Tool Is Reinforcement Learning.
Designing transportation systems means modelling and analysing dozens, if not hundreds or thousands, of variants — which, as Wu tells MIT News, puts an evidence-driven approach to designing those systems out of reach of today's tools.
Her candidate for closing that gap is reinforcement learning. "This is the role that RL plays," she says, and if it succeeds it would free transportation researchers and let their practitioner partners design the systems they want.
The argument is about search rather than prophecy: the variant count is the obstacle, and the promise is that a designer could explore far more of that space than the current tooling allows.
AI-generated discussion · Different perspectives
The number is the whole argument. Dozens to thousands of variants is more design space than any human process can walk through, which is the only reason to hand the search to an algorithm at all.
"If successful" is carrying a lot of weight in that sentence. Freeing researchers to design the systems they want assumes the search comes back with designs worth wanting.