This paper presents an interaction-aware, modular framework for local trajectory planning in autonomous driving, particularly suited for multi-agent racing scenarios. Our framework first identifies viable drivable areas (tunnels), taking into account predictions of other agents’ behaviors, and subsequently utilizes a high-level decision-making module to select the optimal corridor considering both static and moving vehicles. This decision-making module also strategically determines when to follow an opponent or initiate an overtaking maneuver, while ensuring compliance with racing regulations. A Model Predictive Control (MPC) module is then employed to compute an optimal, collision-free trajectory within the chosen corridor. The proposed modular architecture simplifies the computational complexity typically associated with MPC optimization and facilitates independent component testing. Simulations and real-world tests on various racing tracks demonstrate the efficacy of our approach, even in highly dynamic interactive scenarios with multiple simultaneous opponents; videos of these and additional experiments are available at https://atoschi.github.io/tunnels-framework/.

The sideslip angle, crucial for vehicle safety and stability, is determined using both longitudinal and lateral velocities. However, measuring the lateral component often necessitates costly sensors, leading to its common estimation, a topic thoroughly explored in existing literature. This paper introduces LOP-UKF, a novel method for estimating vehicle lateral velocity by integrating Lidar Odometry with the Pacejka tire model predictions, resulting in a robust estimation via an Unscendent Kalman Filter (UKF). This combination represents a distinct alternative to more traditional methodologies, resulting in a reliable solution also in edge cases. We present experimental results obtained using the Dallara AV-21 across diverse circuits and track conditions, demonstrating the effectiveness of our method.

The great quest for adopting AI-based computation for safety-/mission-critical applications motivates the interest towards methods for assessing the robustness of the application w.r.t. not only its training/tuning but also errors due to faults, in particular soft errors, affecting the underlying hardware. Two strategies exist: architecture-level fault injection and application-level functional error simulation. We present a framework for the reliability analysis of Convolutional Neural Networks (CNNs) via an error simulation engine that exploits a set of validated error models extracted from a detailed fault injection campaign. These error models are defined based on the corruption patterns of the output of the CNN operators induced by faults and bridge the gap between fault injection and error simulation, exploiting the advantages of both approaches. We compared our methodology against SASSIFI for the accuracy of functional error simulation w.r.t. fault injection, and against TensorFI in terms of speedup for the error simulation strategy. Experimental results show that our methodology achieves about 99% accuracy of the fault effects w.r.t. SASSIFI, and a speedup ranging from 44x up to 63x w.r.t. TensorFI, that only implements a limited set of error models.