"Friction between the road and vehicle tires plays a key role in defining how a vehicle should be controlled and maneuvered in winter conditions. To enable safe automated driving in winter, our research group is developing computer vision-based methodology for estimating the friction properties of the road. This is achieved with deep learning techniques."  Ojala Risto

”Storing electrical energy from renewable sources onboard the ship represents a promising zero-emission pathway for coastal maritime shipping. The key technical challenge for battery-electric ships arises from the extra volume and weight of the battery system, propulsion motor, and power electronics relative to combustion engines and tanks of conventional ships. Our group is developing modeling techniques to accelerate the design space exploration of novel battery-electric ship concepts. These techniques, known as convex transformations, can accelerate solution times of design problems from hours or days to only a few seconds.” Ritari Antti

"In the field of autonomous driving, targeted external perception is a key component in executing optimum control strategies to ensure the safety of vulnerable road users. The research focuses on developing an attention mechanism in the form of utilizing scene-flows to obtain moving road users in the neighborhood of an ego-vehicle in different driving scenarios. The work revolves on the underlying concepts of sensor integration, dynamic background refinement, pattern analysis and vision-based prediction models."   Jayawickrama Nilusha

"I have been a doctoral candidate in the autonomous mobility laboratory since early 2022. My research focuses on autonomous driving in adverse conditions. I specialize in solutions to mitigate weather effects such as snowfall on sensor data using deep learning tools"  Alvari Seppänen

"Detection of the drivable area in all driving conditions is extremely important for autonomous vehicles and advanced driver assistance systems. My research concentrates on drivable area detection in demanding winter driving conditions with deep learning methods. For better scalability, I try to find ways to automatize the learning process that usually requires human supervision." Alamikkotervo Eerik

Mapping in dynamic environments, produces dynamic points, adversely effecting localization. This study aims to remove the dynamic points by leveraging the observation that stationary points, over multiple scans, have a smaller convex hull volume compared to dynamic ones. Habibiroudkenar Pejman 

"Vehicular energy-efficiency plays a pivotal role in reducing the carbon footprint in the world, especially for long-haul vehicles. To achieve such energy-efficiency in hybrid vehicles, our research group is focused on developing robust controllers which decide when to use internal combustion engines and when to use electric motors fitted with electric batteries in such vehicles. The robust controllers are developed using model-based and convex optimization techniques."  Sahoo Subhadyuti

"Accurate localization is crucial for indoor mobile robots to perform tasks effectively. However, achieving this level of precision is challenging in real-world scenarios due to the inherent complexity of environments and the variable quality of sensor measurements. My research aims to enable reliable and precise localization by quantifying the uncertainty in lidar scan-matching, through the comparison of solid planar features—such as walls and planes—between the map and the current scan". Hari Prasanth

"I'm currently researching testing methods for electric powertrains, such as hardware-in-the-loop simulation. The goal is to accurately emulate the behavior of the final powertrain without needing to test with a real vehicle. Achieving this objective would save development resources and minimize risks associated with costly vehicle testing." Mäkelä Kaarlo

"Road detection in snowy conditions is crucial for safe autonomous driving during winter. In my research, I develop fully automated ways to train road detection models for snowy conditions. In my latest work, I combined the accurate 3d measurements from lidar with camera to automatically generate accurate training labels without any manual  work. " Eerik Alamikkotervo

"Multi-Source Vehicle Tracking by Fusing Connected Vehicle and Static Radar Data

Intelligent transportation systems aim to enhance safety, reduce time spent in traffic and enable vehicular automation. The core of such systems is based of V2X (vehicle-to-everything) communications. In this research the situational awareness of traffic signal control is enhanced by a research vehicle sending lidar based detections of other vehicles via V2X. Theses detections are combined with with static radar detections from the same area. The system utilizes high level data fusion to combat occlusion, and enhances the total detection accuracy in comparison to either data sources by themselves."  Pippuri Aleksi

"My current research addresses the domain of perception in autonomous driving with unsupervised computer vision. The research leverages slot-attention based object-centric learning for improving entity prioritization based on the cues of feature similarities, area, proximity and motion. This work is useful because it drives the capability of autonomous vehicles to intelligently decompose and generalize scenes across various driving scenarios in an unsupervised setting. Segmentation is therefore emergent as it doesn't rely on preset categories for classification, making strides towards mimicking human inference." Nilusha Jayawickrama

“Deep learning -based metric monocular depth estimation methods can be used to generate 3D point clouds similar to those of LiDAR sensors. In my research, I use those depth estimation methods to simulate LiDAR sensors so that we can get better adaptability for mapping with different sensors, add point clouds to existing datasets that only have camera data, and add a secondary 3D view for datasets with only LiDAR.” Eelis Peltola

"Autonomous operation of mobile robots in large-scale and salient free (lack of distinctive features) environments like CERN's facilities present unique localization and mapping difficulties, particularly due to the use of non-repetitive scanning LiDAR systems . Traditional LiDAR registration methods, optimized for repetitive scan patterns, often fail when applied to sparse or irregular data where scan-to-scan overlap is limited. This paper proposes a novel spatial-context-aware LiDAR registration framework designed to improve generalized ICP registeration technique under such non-repetitive scan constraints for mapping CERN facilities." Pejman

"The research leverages infrared and RGB camera frames to perform automated labeling for low-light pedestrian detection tasks, the generated labels can be used directly for model training omitting the need for human labor in the labeling process" Dimitrios Bouzoulas