In this paper we introduce a general estimation methodology for learning a model of human perception and control in a sensorimotor control task based upon a finite set of demonstrations. The model’s structure consists of (i) the agent’s internal representation of how the environment and associated observations evolve as a result of control actions and (ii) the agent’s preferences over observable outcomes. We consider a model’s structure specification consistent with active inference, a theory of human perception and behavior from cognitive science. According to active inference, the agent acts upon the world so as to minimize surprise defined as a measure of the extent to which an agent’s current sensory observations differ from its preferred sensory observations. We propose a bi-level optimization approach to estimation which relies on a structural assumption on prior distributions that parameterize the statistical accuracy of the human agent’s model of the environment. To illustrate the proposed methodology, we present the estimation of a model for car-following behavior based upon a naturalistic dataset. Overall, the results indicate that learning active inference models of human perception and control from data is a promising alternative to closed-box models of driving.

Balancing performance and safety is crucial to deploying autonomous vehicles in multi-agent environments. In particular, autonomous racing is a domain that penalizes safe but conservative policies, highlighting the need for robust, adaptive strategies. Current approaches either make simplifying assumptions about other agents or lack robust mechanisms for online adaptation. This work makes algorithmic contributions to both challenges. First, to generate a realistic, diverse set of opponents, we develop a novel method for self-play based on replica-exchange Markov chain Monte Carlo. Second, we propose a distributionally robust bandit optimization procedure that adaptively adjusts risk aversion relative to uncertainty in beliefs about opponents’ behaviors. We rigorously quantify the tradeoffs in performance and robustness when approximating these computations in real-time motion-planning, and we demonstrate our methods experimentally on autonomous vehicles that achieve scaled speeds comparable to Formula One racecars.

TUNERCAR is a toolchain that jointly optimizes racing strategy, planning methods, control algorithms, and vehicle parameters for an autonomous racecar. In this paper, we detail the target hardware, software, simulators, and systems infrastructure for this toolchain. Our methodology employs a parallel implementation of CMA-ES which enables simulations to proceed 6 times faster than real-world rollouts. We show our approach can reduce the lap times in autonomous racing, given a fixed computational budget. For all tested tracks, our method provides the lowest lap time, and relative improvements in lap time between 7-21%. We demonstrate improvements over a naive random search method with equivalent computational budget of over 15 seconds/lap, and improvements over expert solutions of over 2 seconds/lap. We further compare the performance of our method against hand-tuned solutions submitted by over 30 international teams, comprised of graduate students working in the field of autonomous vehicles. Finally, we discuss the effectiveness of utilizing an online planning mechanism to reduce the reality gap between our simulation and actual tests.

Learning-based methodologies increasingly find applications in safety-critical domains like autonomous driving and medical robotics. Due to the rare nature of dangerous events, real-world testing is prohibitively expensive and unscalable. In this work, we employ a probabilistic approach to safety evaluation in simulation, where we are concerned with computing the probability of dangerous events. We develop a novel rare-event simulation method that combines exploration, exploitation, and optimization techniques to find failure modes and estimate their rate of occurrence. We provide rigorous guarantees for the performance of our method in terms of both statistical and computational efficiency. Finally, we demonstrate the efficacy of our approach on a variety of scenarios, illustrating its usefulness as a tool for rapid sensitivity analysis and model comparison that are essential to developing and testing safety-critical autonomous systems.

While recent developments in autonomous vehicle (AV) technology highlight substantial progress, we lack tools for rigorous and scalable testing. Real-world testing, the de facto evaluation environment, places the public in danger, and, due to the rare nature of accidents, will require billions of miles in order to statistically validate performance claims. We implement a simulation framework that can test an entire modern autonomous driving system, including, in particular, systems that employ deep-learning perception and control algorithms. Using adaptive importance-sampling methods to accelerate rare-event probability evaluation, we estimate the probability of an accident under a base distribution governing standard traffic behavior. We demonstrate our framework on a highway scenario, accelerating system evaluation by 2-20 times over naive Monte Carlo sampling methods and 10-300P times (where P is the number of processors) over real-world testing.

Relaxed notions of decidability widen the scope of automatic verification of hybrid systems. In quasi-decidability and delta-decidability, the fundamental compromise is that if we are willing to accept a slight error in the algorithm's answer, or a slight restriction on the class of problems we verify, then it is possible to obtain practically useful answers. This paper explores the connections between relaxed decidability and the robust semantics of Metric Temporal Logic formulas. It establishes a formal equivalence between the robustness degree of MTL specifications, and the imprecision parameter delta used in delta-decidability when it is used to verify MTL properties. We present an application of this result in the form of an algorithm that generates new constraints to the delta-decision procedure from falsification runs, which can speed up the verification run. We then establish new conditions under which robust testing, based on the robust semantics of MTL, is in fact a quasi-semidecision procedure. These results allow us to delimit what is possible with fast, robustness-based methods, accelerate (near-)exhaustive verification, and further bridge the gap between verification and simulation.