Adequately exercising the behaviors of autonomous vehicles is fundamental to their validation. However, quantifying an autonomous vehicle’s testing adequacy is challenging as the system’s behavior is influenced both by its state as well as its physical environment. To address this challenge, our work builds on two insights. First, data sensed by an autonomous vehicle provides a unique spatial signature of the physical environment inputs. Second, given the vehicle’s current state, inputs residing outside the autonomous vehicle’s physically reachable regions are less relevant to its behavior. Building on those insights, we introduce an abstraction that enables the computation of a physical environment-state coverage metric, PhysCov. The abstraction combines the sensor readings with a physical reachability analysis based on the vehicle’s state and dynamics to determine the region of the environment that may affect the autonomous vehicle. It then characterizes that region through a parameterizable geometric approximation that can trade quality for cost. Tests with the same characterizations are deemed to have had similar internal states and exposed to similar environments and thus likely to exercise the same set of behaviors, while tests with distinct characterizations will increase PhysCov. A study on two simulated and one real system’s dataset examines PhysCovs’s ability to quantify an autonomous vehicle’s test suite, showcases its characterization cost and precision, investigates its correlation with failures found and potential for test selection, and assesses its ability to distinguish among real-world scenarios.

Adversarial test generation techniques aim to produce input perturbations that cause a DNN to compute incorrect outputs. For autonomous vehicles driven by a DNN, however, the effect of such perturbations are attenuated by other parts of the system and are less effective as vehicle state evolves. In this work we argue that for adversarial testing perturbations to be effective on autonomous vehicles, they must account for the subtle interplay between the DNN and vehicle states. Building on that insight, we develop DeepManeuver, an automated framework that interleaves adversarial test generation with vehicle trajectory physics simulation. Thus, as the vehicle moves along a trajectory, DeepManeuver enables the refinement of candidate perturbations to: (1) account for changes in the state of the vehicle that may affect how the perturbation is perceived by the system; (2) retain the effect of the perturbation on previous states so that the current state is still reachable and past trajectory is preserved; and (3) result in multi-target maneuvers that require fulfillment of vehicle state sequences (e.g. reaching locations in a road to navigate a tight turn). Our assessment reveals that DeepManeuver can generate perturbations to force maneuvers more effectively and consistently than state-of-the-art techniques by 20.7 percentage points on average. We also show DeepManeuver's effectiveness at disrupting vehicle behavior to achieve multi-target maneuvers with a minimum 52% rate of success.

Openpilot is an open source system to assist drivers by providing features like automated lane centering and adaptive cruise control. Like most systems for autonomous vehicles, Openpilot relies on a sophisticated deep neural network (DNN) to provide its functionality, one that is susceptible to safety property violations that can lead to crashes. To uncover such potential violations before deployment, we investigate the use of falsification, a form of directed testing that analyzes a DNN to generate an input that will cause a safety property violation. Specifically, we explore the application of a state-of-the-art falsifier to the DNN used in OpenPilot, which reflects recent trends in network design. Our investigation reveals the challenges in applying such falsifiers to real-world DNNs, conveys our engineering efforts to overcome such challenges, and showcases the potential of falsifiers to detect property violations and provide meaningful counterexamples. Finally, we summarize the lessons learned as well as the pending challenges for falsifiers to realize their potential on systems like OpenPilot.

Complex environments can cause robots to fail. Identifying the key elements of the environment associated with such failures is critical for faster fault isolation and, ultimately, debugging those failures. In this work we present the first automated approach for reducing the environment in which a robot failed. Similar to software debugging techniques, our approach systematically performs a partition of the environment space causing a failure, executes the robot in each partition containing a reduced environment, and further partitions reduced environments that still lead to a failure. The technique is novel in the spatial-temporal partition strategies it employs, and in how it manages the potential different robot behaviors occurring under the same environments. Our study of a ground robot on three failure scenarios finds that environment reductions of over 95% are achievable within a 2-hour window.