A map-guided closed-form framework with Riccati-based tracking for low-speed mixed-traffic obstacle avoidance

Engineering notes

Across 60 main trials, the proposed framework achieves a success rate of 1.0 with zero collisions, a mean cross-track error of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mn>0</mml:mn> <mml:mo>.</mml:mo> <mml:mn>021</mml:mn> <mml:mspace width="0.25em"/> <mml:mi mathvariant="normal">m</mml:mi> </mml:mrow> </mml:math> , and a maximum cross-track error of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mn>0</mml:mn> <mml:mo>.</mml:mo> <mml:mn>178</mml:mn> <mml:mspace width="0.25em"/> <mml:mi mathvariant="normal">m</mml:mi> </mml:mrow> </mml:math> . The results support the framework as a reproducible low-speed benchmark and clarify its operational boundaries rather than claiming general validity for high-speed, dense, or strongly interactive traffic.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Low-speed urban encounters with vehicles, bicycles, and pedestrians impose coupled safety and tracking-accuracy requirements on autonomous vehicles. This paper presents a map-guided obstacle-avoidance framework for low-density and weak-interaction mixed-traffic scenarios. The framework integrates an HD-map lane-feasibility decision layer, a closed-form reference-path generator, and a Riccati-recursion-based linear time-varying tracking controller. The decision layer selects a feasible adjacent driving lane from the CARLA waypoint graph, while the path layer synthesises an obstacle-aware reference through sigmoid blending, a fade-in factor that removes spawn-time cross-track artefacts, and a lateral-shift clamp that prevents unrealistic detours. The tracking layer solves the finite-horizon quadratic tracking problem analytically and applies actuator saturation to the resulting commands; it is therefore not claimed to provide the feasibility guarantees of a constrained quadratic-programming MPC. Evaluation is conducted in CARLA 0.9.16 on Town10HD_Opt at target speeds no higher than <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mn>8</mml:mn> <mml:mspace width="0.25em"/> <mml:mi mathvariant="normal">m</mml:mi> <mml:mo stretchy="false">/</mml:mo> <mml:mi mathvariant="normal">s</mml:mi> </mml:mrow> </mml:math> , using five vehicle, bicycle, and pedestrian profiles, four primary baselines, two ablation variants, an adaptive-MPC proxy, a Frenet-quintic planner baseline, and sensitivity/intensity stress tests. Across 60 main trials, the proposed framework achieves a success rate of 1.0 with zero collisions, a mean cross-track error of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mn>0</mml:mn> <mml:mo>.</mml:mo> <mml:mn>021</mml:mn> <mml:mspace width="0.25em"/> <mml:mi mathvariant="normal">m</mml:mi> </mml:mrow> </mml:math> , and a maximum cross-track error of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mn>0</mml:mn> <mml:mo>.</mml:mo> <mml:mn>178</mml:mn> <mml:mspace width="0.25em"/> <mml:mi mathvariant="normal">m</mml:mi> </mml:mrow> </mml:math> . The results support the framework as a reproducible low-speed benchmark and clarify its operational boundaries rather than claiming general validity for high-speed, dense, or strongly interactive traffic.

Links and sources

Comments