Contingencies from Observations: Tractable Contingency Planning with Learned Behavior Models

Humans have a remarkable ability to make decisions by accurately reasoning\nabout future events, including the future behaviors and states of mind of other\nagents. Consider driving a car through a busy intersection: it is necessary to\nreason about the physics of the vehicle, the intentions of other drivers, and\ntheir beliefs about your own intentions. If you signal a turn, another driver\nmight yield to you, or if you enter the passing lane, another driver might\ndecelerate to give you room to merge in front. Competent drivers must plan how\nthey can safely react to a variety of potential future behaviors of other\nagents before they make their next move. This requires contingency planning:\nexplicitly planning a set of conditional actions that depend on the stochastic\noutcome of future events. In this work, we develop a general-purpose\ncontingency planner that is learned end-to-end using high-dimensional scene\nobservations and low-dimensional behavioral observations. We use a conditional\nautoregressive flow model to create a compact contingency planning space, and\nshow how this model can tractably learn contingencies from behavioral\nobservations. We developed a closed-loop control benchmark of realistic\nmulti-agent scenarios in a driving simulator (CARLA), on which we compare our\nmethod to various noncontingent methods that reason about multi-agent future\nbehavior, including several state-of-the-art deep learning-based planning\napproaches. We illustrate that these noncontingent planning methods\nfundamentally fail on this benchmark, and find that our deep contingency\nplanning method achieves significantly superior performance. Code to run our\nbenchmark and reproduce our results is available at\nhttps://sites.google.com/view/contingency-planning\n