Till KTH:s startsida Till KTH:s startsida

Visa version

Version skapad av Patric Jensfelt 2017-11-24 17:05

Visa < föregående | nästa >
Jämför < föregående | nästa >

Master thesis proposals - external

Ericsson

https://jobs.ericsson.com/job/Stockholm-Master-Thesis-on-Immersive-Augmented-Reality-Device-Technologies-AB/441519100/?locale=en_US

Geistt

Using AI methods to Optimize Performance of a Human-UAV Team Performing Humanitarian Disaster Response Delivery of Food and Medicine
Contact: Petter Ögren (petter@kth.se)

Company: Geistt (http://www.geistt.com

In a disaster scenario with severely damaged infrastructure, such as flooding or earthquakes, the delivery of food and medicine is often very difficult.In particular, the so-called last mile logistics, getting items from airports or distribution centers to the victims, is a challenge.However, it is believed that a fleet of small (< 1m) UAVs can contribute in this area. In this thesis project, you will use Unity3D (a game development platform) to support the creation (an initial environment will be available) of a realistic simulation environment to evaluate your AI solutions. The focus of the thesis will be ondesigning and implementing AI capabilities for the UAV system, where example behaviors that are needed include searching for victims, delivering packages to victims, resource/task allocation to the UAV fleet of 3-30 vehicles (who searches where, who delivers what, trade-offs between risk and mission completion etc). The intended system is supposed to be operated by a human rescue worker. However, this person cannot control 30 UAVs at the same time, so an important part of the thesis is to determine an efficient division of work between AI and the human operator. Different design choices will be evaluated using the game functionality of the Unity simulation engine. 

Research questions

  • What AI methods are efficient in a UAV disaster response system?
  • What is a good division of work between Human and system in a UAV disaster response system?

NCC

Computer Vision for increased information quality in MassControl

Contact: Patric Jensfelt (patric@kth.se)

NCC is one of the largest construction companies in the Nordic region. NCC sees great value in reviewing their processes and investigating how digitization can be used to create a better work environment and more effective processes.The information that is fed back to the logistics system / project management for further planning and follow-up is today very limited. This means that it takes time to detect errors and that there is uncertainty about what is performed and when / how.

In this thesis project, we study how to improve the process of loading, transporting and unloading materials by using computer vision to understand which material is on the truck bed and, with relatively good precision, also what volume.Today, an App is used by which the machine operator signals when he / she commences a transport with a certain kind of goods, such as sand. It would be of great value to get more information in the logistics system and to get this information automatically. Examples of the type of information desired are volume, where excavator buckets load the material in relation to the truck and at what time the bucket releases the material on the bed. In the long term, one would also like to decide what type of material is on the bed, so that you can distinguish between sand and soil / rock, etc.In a first step the student will investigate, in cooperation with NCC, more specifically what kind of information is of interest and make a specification. In the next phase, the student will survey previous work in related areas, both in research and industry, and both regarding sensors and algorithms. The initial specification may be adjusted to consider this new knowledge. Based on established methods and what has been developed in the study of previous work, methods will be developed for the work and these should be implemented and evaluated. The aim is to get a better understanding of possibilities and limitations on the subject and to develop simple demonstration.

Scania

Tillämpningsområden för Google Glass inom Scania

Contact: Bashar Mengana, 0736-686941, bashar.mengana@scania.com (skicka ansökan till gunilla.abrink@scania.com, märk ansökan med FU14-140)

Bakgrund:

Vi är på väg mot det uppkopplade samhället. Scania började sin resa 1999 och tog ett stort steg 2011 då alla nya fordon började utrustas med en telematikuppkoppling. Idag har Scania mer än 100 000 uppkopplade fordon vilka tillgängliggör stora mängder fordons- och förardata samt skapar en grund för nya tjänster och arbetssätt.

Det uppkopplade samhället vilar på flera hörnstenar: uppkopplade ting (IoT), molnteknologi, säkerhet, trådlösteknik nya affärsmodeller, tjänster och arbetsmetoder.
Parallellt med detta har wearables (kroppsnära teknik) utvecklats och förväntas bli en viktig ingrediens i det uppkopplade samhället. Wearables kan bidra till att få in praktisk teknik i vardagslivet på ett sådant sätt att det stöttar och effektiviserar arbetsflöden med visuellt och audiellt stöd och bidrar till att koppla upp människan bortom mobiltelefonen.

Mål:

Målet med examensarbetet är reda ut samt ge förslag på vilka tillämpningsområden Google Glass kan bidra till inom Scanias tjänsteutveckling för interaktionen mellan förare, lastbil och åkeri, för att på så sätt öka kunskapsnivån inom området wearables; samt att därefter realisera tillämpning i form av en prototyp (med Google Glass) med syfte att verifiera applikation i verkligheten.

Uppdragsbeskrivning:

I arbetet ingår i huvudsyfte att reda ut hur Google Glass kan bidra till att fördjupa interaktionen mellan fordon – åkeri, genom att förbättra en befintlig Scaniatjänst alternativt tänka utanför ramarna.

Uppdraget omfattar att

  1. sätta sig in i Scanias befintliga affär inom tjänsteutveckling och lära känna Google Glass,

  2. reda ut hur utvalda Scaniatjänster inom befintlig produktportfölj kan förbättras med hjälp av Google Glass,

  3. forska – teoretiskt – fram nya tjänster, byggda på Google Glass plattformen, relevanta för Scanias tjänsteaffär,

  4. välja ut lämplig förbättring av tjänst alternativt ny tjänst att realisera på Google Glass-plattformen, vilket då skall syfta till att illustrera hur wearables kan tillämpas inom Scanias tjänsteutveckling.

Lämplig bakgrund:

Uppdraget kräver två studenter som tillsammans kan utforska behoven samt visualisera idéerna genom att bygga en prototyp. Vi tror att ni är två drivna teknologer inom datateknik, teknisk fysik, industriell ekonomi, elektroteknik (med inriktning mot datalogi) eller liknande. Ni bör ha ett brinna intresse för MDI (människa-datorinteraktion) samt en viss känsla för grafisk form. Kunskap inom Java, Eclipse IDE fodras; Kunskap inom Android SDK och GDK är meriterande. Vi kan hjälpa till att matcha rätt kandidater.

Spotify

Machine learning applied to music search

There are several parts in our search platform that use machine learning to give the best user experience possible. We want a master’s student to look into the best way of doing this.

Machine Learning at Spotify

Computer Science, Data and Security

Volumental

Efficient object segmentation on mobile phones

Since AlexNet published in 2012, Convolutional Neural Networks has ushered a new era in computer vision, consistently improving object detection and segmentation accuracy. In image segmentation, the latest promising work on this front is Mask R-CNN, a region proposing network for object segmentation, building upon a series of CNNs for object detection [1]. This Msc thesis is about implementing Mask R-CNN that can run on flagship iPhone with the end goal of 3D scanning human bodies. As such, the thesis combines theoretical understanding of CNNs with the practice of running it on mobile devices. This is a challenging and exciting thesis topic and we are looking for ambitious, talented candidates.

About Volumental
Volumental is a computer vision company from RPL, KTH active in 3D body scanning and product recommendation based on 3D measurements in footwear. Today we have our computer vision systems are deployed across 32 countries and scanning hundreds of thousands of people regularly, working with some of the world's biggest brands. We are a relatively small but growing team of PhDs in computer vision and machine learning and are product RPL-alumni. We have a healthy gender ratio of almost half of the team being women, and we hail from 9 different countries. We are located centrally at T-Centralen.

If you are interested reach out to: alper@volumental.com

[1] blog.athelas.com/a-brief-history-of-cnns-in-image-segmentation-from-r-cnn-to-mask-r-cnn-34ea