Taxi

The 'Taxi' dataset is a benchmark that contains high-dimensional observations of a grid-based environment where an agent must navigate to pick up and drop off passengers, and it is used to evaluate algorithms for uncovering independently controllable latent variables in reinforcement learning.

Papers using Taxi (3)

From Pixels to Factors: Learning Independently Controllable State Variables for Reinforcement Learning2025Active Query Selection for Crowd-Based Reinforcement Learning2025Explaining Reinforcement Learning: A Counterfactual Shapley Values Approach2024