Search

This master's thesis studies the implementation of advanced machine learning (ML) techniques in industrial automation systems, focusing on applying machine learning to enable and evolve autonomous sorting capabilities in robotic manipulators. In particular, Inverse Kinematics (IK) and Reinforcement Learning (RL) are investigated as methods for controlling a UR10e robotic arm for pick-and-place of moving objects on a conveyor belt within a small-scale sorting facility. A camera-based computer vision system applying YOLOv8 is used for real-time object detection and instance segmentation. Perception data is utilized to ascertain optimal grip points, specifically through an implemented algorithm that outputs optimal grip position, angle, and width. As the implemented system includes testing and evaluation on a physical system, the intricacies of hardware control, specifically the reverse engineering of an OnRobot RG6 gripper is elaborated as part of this study. The system is implemented on the Robotic Operating System (ROS), and its design is in particular driven by high modularity and scalability in mind. The camera-based vision system serves as the primary input, while the robot control is the output. The implemented system design allows for the evaluation of motion control employing both IK and RL. Computation of IK is conducted via MoveIt2, while the RL model is trained and computed in NVIDIA Isaac Sim. The high-level control of the robotic manipulator was accomplished with use of Proximal Policy Optimization (PPO). The main result of the research is a novel reward function for the pick-and-place operation that takes into account distance and orientation from the target object. In addition, the provided system administers task control by independently initializing pick-and-place operation phases for each environment. The findings demonstrate that PPO was able to significantly enhance the velocity, accuracy, and adaptability of industrial automation. Our research shows that accurate control of the robot arm can be reached by training the PPO Model purely by applying a digital twin simulation.

International audience; Macrophages are key cells of innate immune response and serve as the first line of defense against bacteria. Transcription profiling of bacteria-infected macrophages could provide important insights on the pathogenicity and host defense mechanisms during infection. We have examined transcription profiles of bovine monocyte-derived macrophages (bMDMs) isolated from the blood of 12 animals and infected in vitro with two strains of Streptococcus agalactiae. Illumina sequencing of RNA from 36 bMDMs cultures exposed in vitro to either one of two sequence types of S. agalactiae (ST103 or ST12) for 6 h and unchallenged controls was performed. Analyses of over 1,656 million high-quality paired-end sequence reads revealed 5,936 and 6,443 differentially expressed genes (p < 0.05) in bMDMs infected with ST103 and ST12, respectively, versus unchallenged controls. Moreover, 588 genes differentially expressed between bMDMs infected with ST103 versus ST12 were identified. Ingenuity pathway analysis of the differentially up-regulated genes in the bMDMs infected with ST103 revealed significant enrichment for granulocyte adhesion and diapedesis, while significant enrichment for the phagosome formation pathway was found among down-regulated genes. Moreover, Ingenuity pathway analysis of the differentially up-regulated genes in the bMDMs infected with ST12 showed significant enrichment for type 1/type 2 T helper cell activation, while the complement activation pathway was overrepresented in the down-regulated genes. Our study identified pathogen-induced regulation of key genes and pathways involved in the immune response of macrophages against infection but also likely involved in bacterial evasion of the host immune system. These results may contribute to better understanding of the mechanisms underlying subclinical infection such as bovine streptococcal mastitis.

Table S1. RNA samples that were used in the study. RIN - RNA integrity number; RT-qPCR - reverse transcription-quantitative PCR; miRNAs seq - microRNA sequencing. X indicates which samples were used in each of the experiments. (DOCX 16 kb)

Figure S1. Experimental design of reverse transcription-quantitative PCR (RT-qPCR) and microRNA sequencing experiments. S. agal: samples of in vitro exposure of blood monocyte-derived macrophages with live Streptococcus agalactiae strain ST103 or strain ST12, respectively; Neg.con: negative control: sample of uninfected blood monocyte-derived macrophages; 6 h – 6 h infection with Streptococcus agalactiae, 2 h – 2 h exposure to LPS. For details on the experimental design, see the Materials and Methods section. (PDF 134 kb)

Figure S2. Principal component analysis of mapped sequence reads. Control represents sequences from non-infected bovine monocyte-derived macrophage libraries, and a) Streptococcus agalactiae strain ST103 infected; b) Streptococcus agalactiae strain ST12 infected; c) LPS-challenged; and d) Streptococcus agalactiae strain ST103 or strain ST12 infected macrophages isolated from the same animals. (PDF 264 kb)

Table S2. List of primers used for reverse transcription-quantitative PCR (RT-qPCR). (DOCX 14 kb)

Additional file 1. Henderson’s rules for the inverse of the numerator relationship matrix when parental relationships are (partly) based on SNPs. Proof that Henderson’s rules hold when parental relationships are based on SNPs.

article to Husdyrforsøksmøtet 2013 (Norwegian Animal Science meeting).(Joint reference population for Nordic Red breeds)

The main objective of the thesis was to investigate the requirements and possibilities for including feed efficiency (FE) in the breeding goal in dairy cattle and hence enable the genetic improvement of feed efficiency. In addition, possible ways to obtain large scale phenotypic data for the genetic improvement of FE were investigated. The data was provided by Norwegian dairy foods company TINE SA (Ås, Norway), NMBU research farm (Ås, Norway), the Norwegian Dairy Herd Recording System (Ås, Norway) and Animal and Grassland Research and Innovation Centre, Teagasc, Moorepark (Co. Cork, Ireland). The data consisted of records from two research farms, the Norwegian dairy herd recording system and mid-infrared (MIR) spectroscopy of milk. In total, data from 160 lactating Norwegian Red dairy cows and 375 lactating Irish Holstein-Friesian dairy cows were used in the thesis, recorded from 2007 to 2015. Individual feed intake (FI), milk yield (MY), concentration of milk, body weight (BW) and milk spectral recordings were included in the dataset. In Paper I, alternative genomic selection (GS) and traditional Best Linear Unbiased Prediction (BLUP) breeding schemes were compared for the genetic improvement of feed efficiency in simulated Norwegian Red dairy cattle populations. The change in genetic gain over time and achievable selection accuracy were studied for MY and residual feed intake (RFI). When contracted test herds, with genotyped and FE recorded cows as a reference population were used, a reference population size of 4,000 new heifers per year was needed to achieve considerable genetic improvement of feed efficiency. With such a reference population it was possible to reach similar selection accuracies of 0.75 for males than when using progeny testing. It was concluded that the use of contracted test herds with additional recordings (e.g. FE) is a viable option for the genetic improvement of such difficult to record traits. In Paper II, MIR spectra of milk was used to predict dry matter intake (DMI) and net energy intake (NEI) in Norwegian Red dairy cows. Leave-one-out cross-validation and external validation were used to develop and validate prediction equations using five different models. Predictions were performed using either partial least squares regression (PLS) or BLUP. When using the PLS method, the greatest accuracy (R) for predicting DMI (0.54) and NEI (0.65) in the external validation dataset was achieved when using both BW and MY as predictors incombination with the MIR spectra. The Best Linear Unbiased Prediction method gave similar accuracies as PLS but the predictions were biased. This study shows that MIR spectral data can be used to predict NEI as a measure of FI in Norwegian Red dairy cattle, and the prediction accuracy can be increased if BW and/or MY is added to the model. In Paper III, milk MIR data from two countries was combined and used to predict NEI in Norwegian Red dairy cows and effective energy intake (EEI) in Holstein-Friesian dairy cows. Split-sample cross-validation and external validation methods were used to develop and validate prediction equations using four different models. Predictions were performed using either PLS regression, multiple linear regression (MLR) or BLUP methods. Best Linear Unbiased Predictions were implemented either as a single trait or a multi-trait method. Using across country spectra, the R of predicting EEI increased by 0.02 units in both the cross-validation and the external validation compared to the model with spectral information within country only. For NEI, the use of across country MIR decreased the prediction accuracy in the cross-validation by 0.02 units and had no effect on R in the external validation. When NEI was predicted using only the MIR spectral information, single trait BLUP method yielded greater accuracy than PLS. For both NEI and EEI, the greatest accuracy of prediction was achieved using across country MIR spectra. This study shows that MIR spectral data from two countries can be combined and used to increase accuracies of predictions of energy intake (EI) as a measure of feed intake in dairy cattle. If sufficient quantity of FE phenotypic data is available, genetic improvement of feed efficiency is possible. MIR spectral data can be used to predict NEI as a measure of feed intake in Norwegian Red dairy cattle. Also, across country MIR spectral data can be used to predict different energy intake traits. Before including any measure of FE in the breeding program, genetic correlations between predicted feed intake, actual feed intake and other performance traits, especially health and fertility traits must be estimated, and taken into consideration. Hovedformålet med avhandlingen var å undersøke kravene og mulighetene for å inkludere fôreffektivitet (FE) i avlsmålet til melkekyr og dermed muliggjøre den genetiske forbedringen av fôreffektivitet. I tillegg ble mulige måter å oppnå storskala fenotypiske data for genetisk forbedring av FE undersøkt. Dataene ble levert av det norske meieriet TINE SA (Ås, Norge), NMBU-forsøksgård (Ås, Norge), Kukontrollen (Ås, Norge) og Animal and Grassland Research and Innovation Centre, Teagasc, Moorepark (Co. Cork, Ireland). Dataene besto av målinger fra to forskningsgårder, kukontrollen og mid-infrarød spekteranalyse (MIR) av melk. Totalt ble det brukt data fra 160 lakterende melkekyr av rasen Norsk Rødt Fe og 375 lakterende Irske Holstein-Frieser melkekyr i avhandlingen, registrert fra 2007 til 2015. Datasettet inkluderer individuell opptak av fôr (FI), melkeytelse (MY), konsentrasjon av melk, kroppsvekt (BW) og melkespekter. I artikkel I ble alternativ genomisk seleksjon (GS) og tradisjonelle Best Linear Unbiased Prediction (BLUP) avlssystemer sammenlignet for den genetiske forbedringen av fôreffektivitet i simulerte Norske Røde melkekyrpopulasjoner. Forandringen i genetisk gevinst over tid og oppnåelig seleksjonsnøyaktighet ble studert for MY og restinntak av fôr (RFI). Når det ble brukt begrensede testbesetninger med kyr som har registrert genotype og FE som referansepopulasjon, var det nødvendig med en referansepopulasjonsstørrelse på 4000 nye kviger per år for å oppnå betydelig genetisk forbedring av fôreffektivitet. Med en slik referansepopulasjon var det mulig å oppnå seleksjonsnøyaktigheter på 0,75 for okser, tilsvarende bruk av avkomstgransking. Det ble konkludert med at bruk av testbesetninger med tilleggsregistreringer (for eksempel FE) er et overkommelig alternativ for den genetiske forbedring av egenskaper som er vanskelige å registrere. I artikkel II ble MIR av melk brukt til å forutsi inntak av tørrstoff (DMI) og netto energiinntak (NEI) i Norsk Rødt Fe. "Holde-en-utenfor" -kryss-validering og ekstern validering ble brukt til å utvikle og validere prediksjonsligninger ved bruk av fem forskjellige modeller. Prediksjoner ble utført ved bruk av enten Partial Least Squares Regression (PLS) eller BLUP metoder. Ved bruk av PLS-metoden ble den største nøyaktigheten (R) for å forutsi DMI (0,54) og NEI (0,65) i det eksterne valideringsdatasettet oppnådd ved bruk av både BW og MY som prediktorer i kombinasjon med MIR-spektrene. Best Linear Unbiased Prediction -metoden ga lignendenøyaktigheter som PLS, men prognosene var partisk. Denne studien viser at MIR-spektraldata kan brukes til å forutsi NEI som et mål for FI i Norsk Rødt Fe, og prediksjonsnøyaktigheten kan økes dersom BW og/eller MY er lagt til modellen. I artikkel III ble MIR av melk fra to land kombinert og brukt til å forutsi NEI i Norsk Rødt Fe og effektivt energiinntak (EEI) i Holstein-Frieser melkekyr. "Split-sample" -kryss-validering og eksterne valideringsmetoder ble brukt til å utvikle og validere prediksjonsligninger ved bruk av fire forskjellige modeller. Forutsigelser ble utført ved bruk av enten PLS-regresjon, multiple lineære regresjon (MLR) eller BLUP-metoder. Best Linear Unbiased Predictions ble implementert enten som enkeltegenskap eller en fler-egenskapsmetode. Ved å bruke spekter på tvers av land, økte R for å forutsi EEI med 0,02 enheter både i kryssvalidering og ekstern validering sammenlignet med modellen med spekterinformasjon bare innen land. For NEI reduserte bruken av MIR på tvers av land prediksjonsnøyaktigheten i kryssvalideringen med 0,02 enheter og hadde ingen effekt i R i den eksterne valideringen. Når NEI var forutsatt bare ved bruk av MIR-spekterinformasjonen, enkeltegenskap BLUP-metode ga større nøyaktighet enn PLS. For både NEI og EEI ble den største nøyaktigheten av prediksjon oppnådd ved bruk av MIR på tvers av land. Denne studien viser at MIR-spektraldata fra to land kan kombineres og brukes til å forutsi energiinntak (EI) som et mål for inntak av fôr i melkekyr. Hvis tilstrekkelig mengde fenotypiske data om FE er tilgjengelige, er genetisk forbedring av fôreffektivitet mulig. MIR-spektraldata kan brukes til å forutsi NEI som et mål for fôrinntaket i Norsk Rødt Fe. Også kan på tvers av land MIR spekterdata brukes til å forutsi forskjellige energiinntaks karakteristikker. Før det inngår noen måling av FE i avlsprogrammet, må genetiske korrelasjoner mellom predikert fôrinntak, faktisk inntak av fôr og andre ytelsesegenskaper, spesielt helse- og fruktbarhetsegenskaper, estimeres og tas i betraktning.