Customizing autonomous vehicles to align with user preferences while ensuring safety may significantly impact their adoption. Collecting user preference data by asking a large number of comparison questions can be demanding. In this work, we use active learning along with temporal logic descriptions of constraints to enable safe learning of preferences with a reduced number of questions. We take a Bayesian inference approach combined with Weighted Signal Temporal Logic (WSTL), resulting in a WSTL formula that can rank signals based on user preferences and be used for correct-and-custom-by-construction control synthesis. Our method is practical for formulas and signals with various complexity since we compute STL-related values offline. We provide an upper bound for the number of answers in disagreement with user answers. We demonstrate the performance of our method both on synthetic data and by human subject experiments in an immersive driving simulator. We consider two driving scenarios, one involving a vehicle approaching a pedestrian crossing and the other with an overtake maneuver. Our results over synthetic experiments with ground truth weight valuation show that our query selection algorithm converges faster than random query selection. Human subject study results show an average agreement of 94% with user answers during training, and 79% during validation (which increases to 86% when restricted to high confidence results).

The Wizard of Oz (WoZ) method is a widely adopted research approach where a human Wizard “role-plays” a not readily available technology and interacts with participants to elicit user behaviors and probe the design space. With the growing ability for modern large language models (LLMs) to role-play, one can apply LLMs as Wizards in WoZ experiments with better scalability and lower cost than the traditional approach. However, methodological guidance on responsibly applying LLMs in WoZ experiments and a systematic evaluation of LLMs’ role-playing ability are lacking. Through two LLM-powered WoZ studies, we take the first step towards identifying an experiment lifecycle for researchers to safely integrate LLMs into WoZ experiments and interpret data generated from settings that involve Wizards role-played by LLMs. We also contribute a heuristic-based evaluation framework that allows the estimation of LLMs’ role-playing ability in WoZ experiments and reveals LLMs’ behavior patterns at scale.

In this work, we propose a safety-guaranteed personalization for autonomous vehicles by incorporating Signal Temporal Logic (STL) into preference learning problem. We propose a new variant of STL called Parametric Weighted Signal Temporal Logic with a new quantitative semantics, namely weighted robustness. Given a set of pairwise preferences, and by using gradient-based optimization methods, we learn a set of valuations for weights that reflect preferences such that preferred ones have greater weighted robustness value than their non-preferred matches. Traditional STL formulas fail to incorporate preferences due its complex nature. Our initial results with data from a human-subject on an intersection with stop sign driving scenario, in which the participant is asked their preferred driving behavior from pairs of vehicle trajectories, indicate that we can learn a new weighted STL formula that captures preferences while also encoding correctness.

In this work we derive a second-order approach to bilevel optimization, a type of mathematical programming in which the solution to a parameterized optimization problem (the “lower” problem) is itself to be optimized (in the “upper” problem) as a function of the parameters. Many existing approaches to bilevel optimization employ first-order sensitivity analysis, based on the implicit function theorem (IFT), for the lower problem to derive a gradient of the lower problem solution with respect to its parameters; this IFT gradient is then used in a first-order optimization method for the upper problem. This paper extends this sensitivity analysis to provide second-order derivative information of the lower problem (which we call the IFT Hessian), enabling the usage of faster-converging second-order optimization methods at the upper level. Our analysis shows that (i) much of the computation already used to produce the IFT gradient can be reused for the IFT Hessian, (ii) errors bounds derived for the IFT gradient readily apply to the IFT Hessian, (iii) computing IFT Hessians can significantly reduce overall computation by extracting more information from each lower level solve. We corroborate our findings and demonstrate the broad range of applications of our method by applying it to problem instances of least squares hyperparameter auto-tuning, multi-class SVM auto-tuning, and inverse optimal control.

ML is being deployed in complex, real-world scenarios where errors have impactful consequences. In these systems, thorough testing of the ML pipelines is critical. A key component in ML deployment pipelines is the curation of labeled training data. Common practice in the ML literature assumes that labels are the ground truth. However, in our experience in a large autonomous vehicle development center, we have found that vendors can often provide erroneous labels, which can lead to downstream safety risks in trained models.

We formulate numerically-robust inductive proof rules for unbounded stability and safety properties of continuous dynamical systems. These induction rules robustify standard notions of Lyapunov functions and barrier certificates so that they can tolerate small numerical errors. In this way, numerically-driven decision procedures can establish a sound and relative-complete proof system for unbounded properties of very general nonlinear systems. We demonstrate the effectiveness of the proposed rules for rigorously verifying unbounded properties of various nonlinear systems, including a challenging powertrain control model.

We develop a set of contracts for autonomous control software that ensures that if all traffic participants follow the contracts, the overall traffic system will be collision-free. We express our contracts in Signal Temporal Logic (STL), a lightweight specification language that enables V &V methodologies. We demonstrate how the specification can be used for evaluation of the performance of autonomy software, and We provide preliminary evidence that our contracts are not excessively conservative, i.e., they are not more restrictive than existing guidelines for safe driving by humans.

This paper proposes a method to evaluate Signal Temporal Logic (STL) robustness formulas using computation graphs. This method results in efficient computations and enables the use of backpropagation for optimizing over STL parameters. Inferring STL formulas from behavior traces can provide powerful insights into complex systems, such as longterm behaviors in time-series data. It can also be used to augment existing prediction and planning architectures by ensuring specifications are met. However, learning STL formulas from data is challenging from a theoretical and numerical standpoint. By evaluating and learning STL formulas using computation graphs, we can leverage the computational efficiency and utility of modern machine learning libraries. The proposed approach is particularly effective for solving parameteric STL (pSTL) problems, the problem of parameter fitting for a given signal. We provide a relaxation technique that makes this method tractable when solving general pSTL formulas. Through a traffic-weaving case-study, we show how the proposed approach is effective in learning pSTL parameters, and how it can be applied for scenario-based testing for autonomous driving and other complex robotic systems.

Deep learning algorithms can fare poorly when the training dataset suffers from heavy class-imbalance but the testing criterion requires good generalization on less frequent classes. We design two novel methods to improve performance in such scenarios. First, we propose a theoretically-principled label-distribution-aware margin (LDAM) loss motivated by minimizing a margin-based generalization bound. This loss replaces the standard cross-entropy objective during training and can be applied with prior strategies for training with class-imbalance such as re-weighting or re-sampling. Second, we propose a simple, yet effective, training schedule that defers re-weighting until after the initial stage, allowing the model to learn an initial representation while avoiding some of the complications associated with re-weighting or re-sampling. We test our methods on several benchmark vision tasks including the real-world imbalanced dataset iNaturalist 2018. Our experiments show that either of these methods alone can already improve over existing techniques and their combination achieves even better performance gains.

We propose a new abstraction refinement procedure based on machine learning to improve the performance of nonlinear constraint solving algorithms on large-scale problems. The proposed approach decomposes the original set of constraints into smaller subsets, and uses learning algorithms to propose sequences of abstractions that take the form of conjunctions of classifiers. The core procedure is a refinement loop that keeps improving the learned results based on counterexamples that are obtained from partial constraints that are easy to solve. Experiments show that the proposed techniques significantly improve the performance of state-of-the-art constraint solvers on many challenging benchmarks. The mechanism is capable of producing intermediate symbolic abstractions that are also important for many applications and for understanding the internal structures of hard constraint solving problems.

Today's automotive industry is making a bold move to equip vehicles with intelligent driver assistance features. A modern automobile is now equipped with a powerful computing platform to run multiple machine learning algorithms for environment perception (e.g., pedestrian detection) and motion control (e.g., vehicle stabilization). These machine learning systems must be highly robust with extremely small failure rate in order to ensure safe and reliable driving. In this paper, we propose a novel Subset Sampling (SUS) algorithm to efficiently validate a machine learning system. In particular, a Markov Chain Monte Carlo algorithm based on graph mapping is developed to accurately estimate the rare failure rate with a minimal amount of test data, thereby minimizing the validation cost. Our numerical experiments show that SUS achieves 15.2× runtime speed-up over the conventional brute-force Monte Carlo method.

Abstract This paper is a tutorial on how to model hybrid systems as hybrid programs in differential dynamic logic and how to prove complex properties about these complex hybrid systems in KeYmaera, an automatic and interactive formal verification tool for hybrid systems. Hybrid systems can model highly nontrivial controllers of physical plants, whose behaviors are often safety critical such as trains, cars, airplanes, or medical devices. Formal methods can help design systems that work correctly. This paper illustrates how KeYmaera can be used to systematically model, validate, and verify hybrid systems. We develop tutorial examples that illustrate challenges arising in many real-world systems. In the context of this tutorial, we identify the impact that modeling decisions have on the suitability of the model for verification purposes. We show how the interactive features of KeYmaera can help users understand their system designs better and prove complex properties for which the automatic prover of KeYmaera still takes an impractical amount of time. We hope this paper is a helpful resource for designers of embedded and cyber–physical systems and that it illustrates how to master common practical challenges in hybrid systems verification.

The use of deductive techniques, such as theorem provers, has several advantages in safety verification of hybrid systems; however, state-of-the-art theorem provers require manual intervention to handle complex systems. Furthermore, there is often a gap between the type of assistance that a theorem prover requires to make progress on a proof task and the assistance that a system designer is able to provide directly. This paper presents an extension to KeYmaera, a deductive verification tool for differential dynamic logic; the new technique allows local reasoning using system designer intuition about performance within particular modes as part of a proof task. Our approach allows the theorem prover to leverage forward invariants, discovered using numerical techniques, as part of a proof of safety. We introduce a new inference rule into the proof calculus of KeYmaera, the forward invariant cut rule, and we present a methodology to discover useful forward invariants, which are then used with the new cut rule to complete verification tasks. We demonstrate how our new approach can be used to complete verification tasks that lie out of the reach of existing automatic verification approaches using several examples, including one involving an automotive powertrain control system.

Lyapunov functions are used to prove stability and to obtain performance bounds on system behaviors for nonlinear and hybrid dynamical systems, but discovering Lyapunov functions is a difficult task in general. We present a technique for discovering Lyapunov functions and barrier certificates for nonlinear and hybrid dynamical systems using a search-based approach. Our approach uses concrete executions, such as those obtained through simulation, to formulate a series of linear programming (LP) optimization problems; the solution to each LP creates a candidate Lyapunov function. Intermediate candidates are iteratively improved using a global optimizer guided by the Lie derivative of the candidate Lyapunov function. The analysis is refined using counterexamples from a Satisfiability Modulo Theories (SMT) solver. When no counterexamples are found, the soundness of the analysis is verified using an arithmetic solver. The technique can be applied to a broad class of nonlinear dynamical systems, including hybrid systems and systems with polynomial and even transcendental dynamics. We present several examples illustrating the efficacy of the technique, including two automotive powertrain control examples.

This paper describes the architecture and implementation of a distributed autonomous gardening system. The garden is a mesh network of robots and plants. The gardening robots are mobile manipulators with an eye-in-hand camera. They are capable of locating plants in the garden, watering them, and locating and grasping fruit. The plants are potted cherry tomatoes enhanced with sensors and computation to monitor their well-being (e.g. soil humidity, state of fruits) and with networking to communicate servicing requests to the robots. Task allocation, sensing and manipulation are distributed in the system and de-centrally coordinated. We describe the architecture of this system and present experimental results for navigation, object recognition and manipulation.