Chalmers Open Digital Repository

Välkommen till Chalmers öppna digitala arkiv!

Här hittar du:

Studentarbeten utgivna på lärosätet, såväl kandidatarbeten som examensarbeten på grund- och masternivå
Digitala specialsamlingar, som t ex Chalmers modellkammare
Utvalda projektrapporter

Forskningspublikationer, rapporter och avhandlingar hittar du i research.chalmers.se

Enheter i Chalmers ODR

Välj en enhet för att se alla samlingar.

Visar 1 - 2 av 2

Senast publicerade

Lattice Surgery med kvantfelskorrektion på IBM Nighthawk r1
(2026) Lilja, Max; Lorentzon, Jacob; Molinder, Ludvig; Nguyen, Jonathan; Nikolic, Minna; Tollander, Gustav
Kvantdatorer har stor potential att utföra särskilt kraftfulla beräkningar för vissa typer av problem, men är i sin nuvarande form bruskänsliga, särskilt när problemstorleken ökar. Därför används kvantfelskorrektion, där kvanttillstånd lagras med extra redundans som logiska tillstånd. Genom upprepade rundor av partiella mätningar kan utfallen avkodas och fel identifieras, utan att det logiska tillståndet kollapsar. I det här arbetet implementerades kvantfelskorrektion med roterade ytkoder, där avkodningen gjordes med MinimumWeight Perfect Matching. Därefter tillämpades lattice surgery, för att flera ytkoder skulle kunna interagera. Ytkoden testades både genom simulering med slumpmässiga injicerade fel och genom att köras på den fysiska kvantprocessorn IBM Nighthawk r1. På kvantprocessorn testades även hur lattice surgery kan användas för att bilda sammanflätade logiska Belltillstånd, och den logiska CNOT-grinden testades genom simulering. Enskilda ytkoder och avkodning på kvantprocessorn resulterade i 5–20% logiska fel för 2–14 rundor, och när lattice surgery användes för att skapa ett Belltillstånd erhölls de teoretiskt förväntade resultaten i över 85% av fallen vid 12 totala rundor. Resultaten tyder på att roterade ytkoder och lattice surgery har god potential att leda till effektiv, feltolerant kvantberäkning i takt med att större kvantprocessorer blir tillgängliga.
AI–baserat träningsstöd för löpning
(2026) Boissier, Christopher; Ivanovic, Cornelia; Johannesson, Fabian; Melin Romstad, Daniel; Tambur, Teodor; Thenander, Ebba
In the field of middle- and long–distance running in athletics, personalized training plans are an important factor for athletic success, yet the creation and adjustment of these plans remain a time–consuming process for coaches. Coaching methodologies often vary based on individual experience and philosophy, placing high demands on the flexibility of digital tools. This thesis explores how advanced AI models can be utilized to automate parts of this decision–making logic, aiming to reduce the administrative workload for coaches and enable an increased focus on technical analysis and active coaching. Through a needs analysis, a training platform has been developed to serve as an AI–based decision support system. The system integrates a regression model for predicting running pace, based on individual Garmin data, with a model based on retrieval–augmented generation for generating structured training schedules from educational material and the athlete’s profile. To enable individualized adjustments, a Large Language Model has been fine-tuned on specific coaching logic to handle athlete’s current physical condition. The result is a web–based prototype demonstrating how AI and machine learning can complement human expertise within middle- and long–distance running.
Real-Time LiDAR Sensor Modeling: Intensity Modeling and Evaluation for Autonomous Vehicle Simulation
(2026) Zhou, Jingbo; Garau Chen, Hailan
Realistic LiDAR simulation is important for the development and validation of autonomous driving systems, but accurately reproducing LiDAR intensity remains challenging. Unlike point geometry, intensity depends on range, incidence angle, surface reflectivity, sensor-specific processing, and environmental effects. In addition, evaluating simulated intensity against real-world data is difficult because exact pointwise alignment between real and simulated point clouds is rarely achievable in a digital twin environment. This thesis investigates LiDAR intensity simulation in a CARLA-based digital twin of the AstaZero proving ground, developed in connection with Volvo Autonomous Solutions. Real-world LiDAR reference data are reconstructed from MCAP recordings and used to evaluate the simulated intensity output. A physically motivated intensity model is introduced for the simulated LiDAR, incorporating the main factors that affect return strength, including range, incidence angle, and material reflectivity. However, because the target LiDAR sensor outputs a vendor-specific value affected by an inaccessible, proprietary internal processing pipeline, a direct analytical sensor model is unattainable. Hence, the framework complements this physical formulation to a final calibrated reflectivity simulation model through empirical distribution mapping. The resulting model serves as a practical, real-time approximation of calibrated reflectivity behavior rather than a complete reproduction of the internal sensor-processing pipeline. To evaluate simulated LiDAR intensity, this thesis combines conventional histogrambased metrics with a novel geometry tolerant evaluation method proposed in this work. Wasserstein distance and Jensen–Shannon distance are used as baseline measures of global intensity distribution agreement. The proposed spherical harmonic based method represents each LiDAR frame as an angular intensity function on the sphere and compares frames using a weighted distance between their degree-wise spherical harmonic energy descriptors. This method captures coarse angular intensity structure in a rotation invariant manner without requiring exact pointwise correspondence. The results show that the proposed intensity model improves the similarity between simulated and real-world reference intensity distributions. The proposed evaluation method also provides a more informative comparison than traditional distributionbased metrics by preserving directional intensity structure when local geometric mismatch is present.
Scalable Vision–Language Machine Learning for Semantic Retrieval of Autonomous Driving Logs
(2026) Albarham, Mohammad; Berggren, Linus
This thesis studies scalable semantic retrieval of autonomous driving multi-view videos recorded with synchronized multi-camera systems. Using a subset of 3,502 multi-view driving videos from NVIDIA’s PhysicalAI Autonomous Vehicles dataset, the work investigates text-to-multi-view video retrieval using natural-language queries and learned cross-modal embeddings. Because the dataset does not contain paired textual descriptions, the proposed pipeline generates pseudo ground-truth captions from sampled video frames using a pretrained vision-language model and extracts frozen text and visual embeddings with jina-clip-v2. These generated captions provide the supervision used for training and evaluation without requiring manual annotation. Lightweight trainable alignment heads are then used to map text and video representations into a shared embedding space, while multi-view representations are constructed through view-level and temporal aggregation. The results quantify the difference between single-view (front camera) and multi-view (front and surrounding cameras) retrieval representations. Extending the representation from a single front-facing camera to six synchronized camera views increases Recall@ 5 from 71% to 85% and Recall@10 from 81% to 93%, indicating improved separation of ground-truth matches in the learned embedding space. In contrast, the LLM-based semantic similarity score changes only marginally, from 77 to 79, suggesting that both retrieval settings often retrieve semantically related driving scenarios. The experiments further show that temporal sampling can be reduced considerably with only minor changes in retrieval performance, indicating substantial redundancy in densely sampled driving video. Since the supervision is derived from automatically generated captions rather than human-annotated descriptions, the retrieval results should be interpreted with that limitation in mind. Overall, the thesis demonstrates that frozen pretrained encoders combined with lightweight fusion and alignment modules provide a computationally scalable approach for semantic retrieval of large-scale autonomous driving multi-view videos.
An FPGA-Based Visual Processing System for Autonomous Driving under Extreme Lighting
(2026) Wu, Chen; Wu, Xingchen
Extreme lighting conditions degrade camera image quality and may cause fatal autonomous driving accidents through perception failures. This thesis presents an FPGA-based front-end visual processing system that mitigates such degradation in real time while preserving downstream model recognition accuracy. The system architecture contributes two innovations: the traditional image signal proces sor pipeline is pruned from over ten hardware modules to five essential modules based on literature consensus, and black-level correction, auto exposure gain, and localized bloom suppression are consolidated into a single Bayer-domain affine correction applied before demosaicing, severing the saturation-energy diffusion cascade at its source. Auto exposure control operates as an independent side channel with zero pixel-throughput overhead. Validation follows a task-driven paradigm using contrastive-learning-based style transfer, where downstream model accuracy rather than pixel-level metrics serves as the evaluation criterion. FPGA implementation achieves low resource consumption, deterministic fixed latency, and throughput far exceeding real-time requirements. Style-transfer-trained models match the accuracy of those trained on original data in both object detection and semantic segmentation tasks, and visual comparison against a professional action camera confirms comparable overall image quality under both daytime and nighttime conditions, while achieving substantially lower and deterministic pipeline latency.