حميد دانشور

A key challenge for advanced path-planning architectures is the loss of resilience an‎d dynamic quality due to occlusion an‎d sensor noise, particularly in complex urban situations. This research designs an‎d eva‎luates an end-to-end path-planning architecture that enables a decision-making core based on the TCP model, with dual-branch (trajectory-control) design, to take advantage of Vehicle-to-Infrastructure (V2I) communication. Motivated by cooperative perception paradigms, the vehicle employs a driving-related communication strategy in which the vehicle creates a "Request Map" based on its predicted path to selec‎tively request pertinent data from roadside units (RSUs). Once aligned, this data is distributed (fused) with the vehicleʹs own LiDAR data in a common Birdʹs-Eye View (BEV) space, employing an attention-based fusion approach, that optimizes ban‎dwidth while developing a rich understan‎ding of the environment. The suggested framework was trained an‎d tested on the V2Xverse dataset, within the CARLA simulator. The performance of the proposed framework was compared against a non-connected baseline, in multiple urban scenarios (left turns, straight driving with occlusions an‎d right turns) under ideal an‎d critical conditions (the vehicle pose was corrupted with Gaussian noise). The results showed a large measure of improvement in the path’s dynamic quality. For example, in the case of left turn, the average maximum curvature decreased by 78/6% (from 47/045 to 10/0663 1/mm) an‎d the yaw rate at reference speed of 5m/s decreased from 13/5 to 3 deg/s. Likewise, for the right turn it was noted the curvature improved by 37% an‎d for urban driving 67% improvement was noted. In critical conditions of 0.2m noise, the connected architecture presented more robustness by decreasing the Average Displacement Error (ADE) by 7% for left turn an‎d 4% for straight driving. The source code an‎d pre-trained models are publicly available at: https://github.com/hamid79daneshvar/CORTEX

خودروي خودران , بهبود مسيريابي , ارتباط خودرو با زيرساخت , دقت مسيريابي , كيفيت ديناميكي مسيريابي , ادراك مشاركتي

Autonomous Driving , Path Planning Improvement , Vehicle-to-Infrastructure (V2I) , Path planning accuracy , dynamic path planning quality , Cooperative Perception

hamid_daneshvar

Dr. Masoud_Masih Tehrani