Machine Learning and Statistical Models to Reduce Perceived Latency of Delay-Sensitive Applications

Aug 5, 2026

Upload by Asher

In an evolving world of communication technology, the demand for low latency and high reliability network architecture continues to rise. Uncontrollable events such as congestion, packet loss, and jitter introduce delays that directly degrade the quality of service for delay sensitive applications. This thesis examines how machine learning can mitigate perceived latency in human-to-machine interaction by forecasting future system instructions. A physical testbed with a robotic arm, remote controller, command interpretation PC, and inference proxy PC supports training, testing, evaluation, and tuning of models.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages59

Reading Time2 h 29 min

AccessFree to download

Problem description

Thesis aim and approach

Forecasting instructions in a robotic context

Abstract and results

Testbed and model evaluation

Sammendrag (Summary)

Preface

Machine Learning and Statistical Models to Reduce Perceived Latency of DelaySensitive Applications Scherman Olsen, Kristoffer Submission date: June 2024 Main supervisor: Lange, Stanislav, NTNU Co-supervisor: Zinner, Thomas, NTNU Norwegian University of Science and Technology Department of Information Security and Communication Technology Title: Machine Learning and Statistical Models to Reduce Perceived La tency of Delay-Sensitive Applications Student: Scherman Olsen, Kristoffer Problem description: Today’s communication heavily relies on the internet. Devices are increasingly getting connected, creating higher quality thresholds for achieving the expected user experience. The best-effort nature of packet delivery over the internet makes it challenging to guarantee consistent, low-latency performance, leading to inconsistent and unpredictable quality in different applications. Hence latency compensation is a crucial piece of future network architectures. This project aims to analyze, evaluate and extend existing solutions for reducing perceived latency by forecasting instructions in a robotic context. This will be achieved by using AI and statistical methods to forecast instructions sent from a remote controller to a robot in an isolated system. Approved on: 2024-02-26 Main supervisor: Lange, Stanislav, NTNU Co-supervisor: Zinner, Thomas, NTNU Abstract In an evolving world of communication technology, the demand for low latency, high reliability network architecture is increasing. Uncontrollable events such as network congestion, packet loss, and jitter can cause delays impacting the quality of service of delay sensitive applications. As Artificial Intelligence (AI) and machine learning models are becoming more popular, they have entered the field of latency mitigation and could be a potential solution to improve existing solutions. This thesis investigates the use of machine learning models to reduce the perceived latency in human to machine interaction. This work focuses primarily on the technical aspects, but is a step towards since it can bea foundation to conduct more evaluations regarding perceived latency. A physical testbed consisting of a robotic arm, a remote controller, a PC for interpreting commands and a proxy-PC to conduct inference when necessary is presented to train, test, evaluate and tune a set of machine learning models that through previous work and research within similar problems have shown promising results. Experimental results indicate that the models are able to predict the future state of the system with relative high accuracy, and that the models are able to reduce the perceived latency within a defined time horizon. Sammendrag I en verden med stadig utvikling innenfor kommunikasjonsteknologi økeretterspørselen etter nettverksarkitektur med lave forsinkelser og høy pålitelighet. Ukontrollerbare hendelser som nettverksbelastning, pakketap og jitter kan forårsake forsinkelser som påvirker kvaliteten på tjenester for forsinkelses-følsomme applikasjoner. Ettersom kunstig intelligens og maskinlæringsmodeller blir mer populære, har de nå kommet inn på området forsinkelses reduksjon og kan være en potensiell løsning for å forbedre eksisterende løsninger. Denne oppgaven undersøker bruken av maskinlæringsmodeller for å redusere oppfattet forsinkelse i interaksjon mellom menneske og maskin. Et fysisk testmiljø bestående av en robotarm, en fjernkontroll, en PC for å tolke kommandoer og en proxy-PC for å utføre inferens når nødvendig, presenteres for å trene, teste, evaluere og justere et sett med maskinlæringsmodeller som gjennom tidligere arbeid og forskning innen lignende problemer har vist lovende resultater. Eksperimentelle resultater indikererat modellene er i stand til å forutsi systemets fremtidige tilstand med relativt høy nøyaktighet, og at modellene er i stand til å redusere den oppfattede forsinkelsen innenfor en definert tidsramme. Preface Writing a thes

FAQs

Why is reducing perceived latency important for delay-sensitive applications?

Internet packet delivery is best-effort, so congestion, packet loss, and jitter can cause inconsistent and unpredictable delays that degrade quality of service for delay sensitive applications.

What is the thesis approach to reduce perceived latency?

The work analyzes existing solutions and extends them by using AI and statistical methods to forecast instructions sent from a remote controller to a robot in an isolated system.

What experimental setup is used to train and evaluate the machine learning models?

The thesis uses a physical testbed including a robotic arm, a remote controller, a PC for interpreting commands, and a proxy-PC that performs inference when necessary.

Aug 5, 2026

Upload by Asher