Counterfactual Simulation
A simulation method that quantifies an agent's actual impact on other members of a system by hypothetically removing that agent and comparing the resulting scenario with the original. The approach is commonly used in causal inference and for reward attribution in multi-agent systems, enabling precise measurement of individual agent contributions to collective outcomes.
Timeline (last 90 days)
Marginal cost is measured by constructing a counterfactual simulation, i.e., hypothetically removing a given AV from the traffic network and comparing the travel times of other vehicles with and without it.
The study proposes computing marginal cost only for AVs whose route choices have the greatest impact on the traffic system, in order to reduce the number of counterfactual simulations and lower computational costs.
The naive approach to counterfactual simulation requires running one complete traffic simulation for each of the N AVs, with computational overhead growing linearly or even super-linearly with fleet size.
All Facts (3)
Marginal cost is measured by constructing a counterfactual simulation, i.e., hypothetically removing a given AV from the traffic network and comparing the travel times of other vehicles with and without it.
50%UnverifiedThe study proposes computing marginal cost only for AVs whose route choices have the greatest impact on the traffic system, in order to reduce the number of counterfactual simulations and lower computational costs.
50%UnverifiedThe naive approach to counterfactual simulation requires running one complete traffic simulation for each of the N AVs, with computational overhead growing linearly or even super-linearly with fleet size.
50%