arXiv subjects

Tor Skeie

Publications and source records attributed to Tor Skeie.

4 recordsLinked to original sources

arXiv2026

GAPL: Grounded Action-effect Policy Learning for LLM-Based Trajectory Planning

Trajectory planning for autonomous driving requires both high-level reasoning and precise low-level control. Large Language Models (LLMs) offer semantic-rich planning capabilities, however, their application is limited by hallucinated reasoning, poor grounding in environment dynamics, and limited numerical precision in control. We propose GAPL (Grounded Action-effect Policy Learning), a unified framework that integrates LLM-based effect estimation, simulation-based effect grounding, and policy optimization into a closed-loop system. GAPL consists of three modules: (1) an LLM-based Effect Evaluator for structured multi-dimensional action-effect estimation; (2) a Simulation-based Effect Grounder that predicts dynamics-consistent effects from simulator rollouts; and (3) an Effect-Aware Decision Maker that grounds LLM effect estimates against simulation via a distiller to guide Proximal Policy Optimization (PPO)-based policy learning. Experiments on four Highway-env scenarios demonstrate that GAPL consistently outperforms baselines, achieving average reductions of {0.76, 0.86, 2.00} in collision rate, average displacement error (ADE), and final displacement error (FDE), and an average reward gain of 1.44.

cs.RO↗

arXiv2026

C-TRAIL: A Commonsense World Framework for Trajectory Planning in Autonomous Driving

Trajectory planning for autonomous driving increasingly leverages large language models (LLMs) for commonsense reasoning, yet LLM outputs are inherently unreliable, posing risks in safety-critical applications. We propose C-TRAIL, a framework built on a Commonsense World that couples LLM-derived commonsense with a trust mechanism to guide trajectory planning. C-TRAIL operates through a closed-loop Recall, Plan, and Update cycle: the Recall module queries an LLM for semantic relations and quantifies their reliability via a dual-trust mechanism; the Plan module injects trust-weighted commonsense into Monte Carlo Tree Search (MCTS) through a Dirichlet trust policy; and the Update module adaptively refines trust scores and policy parameters from environmental feedback. Experiments on four simulated scenarios in Highway-env and two real-world levelXData datasets (highD, rounD) show that C-TRAIL consistently outperforms state-of-the-art baselines, reducing ADE by 40.2%, FDE by 51.7%, and improving SR by 16.9 percentage points on average. The source code is available at https://github.com/ZhihongCui/CTRAIL.

cs.AI↗

arXiv2022

A Lower Bound on Latency Spikes for Capacity-Seeking Network Traffic

Most Internet traffic is carried by capacity-seeking protocols such as TCP and QUIC. Capacity-seeking protocols probe to find the maximum available throughput from sender to receiver, and, once they converge, attempt to keep sending traffic at this maximum rate. Achieving reliable low latency with capacity-seeking end-to-end methods is not yet entirely solved. We contribute a theoretical analysis to this ongoing discussion. In this work, we derive an expression for the minimum size of the spike in latency caused by a sudden drop in network capacity. Our results highlight a quantifiable and fundamental constraint on capacity-seeking network traffic. When end-to-end capacity is suddenly reduced, capacity-seeking traffic inevitably produces a latency spike. A lower bound on this latency spike can be calculated by multiplying the round-trip delay from the network bottleneck to the source of capacity-seeking traffic by the magnitude of the end-to-end capacity reduction. Testbed experiments show that this bound holds for the DCTCP, BBR, and Cubic congestion control algorithms. Our results have implications for the design of low-latency PHY and MAC-layer technologies because we quantify an important transport-layer consequence of unstable traffic rates.

cs.NI↗

arXiv2021

A Model of WiFi Performance With Bounded Latency

In September 2020, the Broadband Forum published a new industry standard for measuring network quality. The standard centers on the notion of quality attenuation. Quality attenuation is a measure of the distribution of latency and packet loss between two points connected by a network path. A vital feature of the quality attenuation idea is that we can express detailed application requirements and network performance measurements in the same mathematical framework. Performance requirements and measurements are both modeled as latency distributions. To the best of our knowledge, existing models of the 802.11 WiFi protocol do not permit the calculation of complete latency distributions without assuming steady-state operation. We present a novel model of the WiFi protocol. Instead of computing throughput numbers from a steady-state analysis of a Markov chain, we explicitly model latency and packet loss. Explicitly modeling latency and loss allows for both transient and steady-state analysis of latency distributions, and we can derive throughput numbers from the latency results. Our model is, therefore, more general than the standard Markov chain methods. We reproduce several known results with this method. Using transient analysis, we derive bounds on WiFi throughput under the requirement that latency and packet loss must be bounded.

cs.NI↗