Contact Info

  E-mail:

  bellonemauro@gmail.com 

  mauro.bellone@taltech.ee

  Skype: bellonemauro

Mauro Bellone: Roboticist, entrepreneur and innovation technology specialist 

About

I received the M.S. degree in Automation Engineering from the University of Salento, Lecce, Italy, where I received the Ph.D. in Mechanical and Industrial Engineering in 2014.

My research interest spans across several areas including artificial intelligence, mobile robotics, autonomous vehicles, energy, computer vision and control systems. My research focuses on the area of advanced sensory perception for mobile robotics and computer vision. 

In 2009, I had the pleasure to visit the Space Robotics Lab of Tohoku University, Sendai, Japan. 

In 2013-14, as visiting researcher at University of Almerìa, I worked on new advanced navigation techniques for mobile robotics and autonomous driving. 

Since 2015 to 2020, I worked with the applied artificial intelligence research group of Chalmers University of Technology, where I actively contributed on several autonomous driving projects. 

In 2021, I was appointed as adjunct professor at Tallinn University of technology supporting the research team of FinEst center and the autonomous vehicle lab in the area of smart transportation systems. 

" we are part of a 10 billion people brain  " 

CLFT: Camera-LiDAR Fusion Transformer for Semantic Segmentation in Autonomous Driving (2024)

Critical research about camera-and-LiDAR-based semantic object segmentation for autonomous driving significantly benefited from the recent development of deep learning. Specifically, the vision transformer is the novel ground-breaker that successfully brought the multi-head-attention mechanism to computer vision applications. Therefore, we propose a vision-transformer-based network to carry out camera-LiDAR fusion for semantic segmentation applied to autonomous driving. Our proposal uses the novel progressive-assemble strategy of vision transformers on a double-direction network and then integrates the results in a cross-fusion strategy over the transformer decoder layers. Unlike other works in the literature, our camera-LiDAR fusion transformers have been evaluated in challenging conditions like rain and low illumination, showing robust performance. The paper reports the segmentation results over the vehicle and human classes in different modalities: camera-only, LiDAR-only, and camera-LiDAR fusion. We perform coherent controlled benchmark experiments of CLFT against other networks that are also designed for semantic segmentation. The experiments aim to evaluate the performance of CLFT independently from two perspectives: multimodal sensor fusion and backbone architectures. The quantitative assessments show our CLFT networks yield an improvement of up to 10% for challenging dark-wet conditions when comparing with Fully-Convolutional-Neural-Network-based (FCN) camera-LiDAR fusion neural network. Contrasting to the network with transformer backbone but using single modality input, the all-around improvement is 5-10%.

Full paper available here and arxiv 

Real-Time Multi-Modal Active Vision for Object Detection on UAVs Equipped With Limited Field of View LiDAR and Camera (2023)

This letter aims to solve the challenging problems in multi-modal active vision for object detection on unmanned aerial vehicles (UAVs) with a monocular camera and a limited Field of View (FoV) LiDAR. The point cloud acquired from the low-cost LiDAR is firstly converted into a 3-channel tensor via motion compensation, accumulation, projection, and up-sampling processes. The generated 3-channel point cloud tensor and RGB image are fused into a 6-channel tensor using an early fusion strategy for object detection based on a Gaussian YOLO network structure. To solve the low computational resource problem and improve the real-time performance, the velocity information of the UAV is further fused with the detection results based on an extended Kalman Filter (EKF). A perception-aware model predictive control (MPC) is designed to achieve active vision on our UAV. According to our performance evaluation, our pre-processing step improves other literature methods running time by a factor of 10 while maintaining acceptable detection performance. Furthermore, our fusion architecture reaches 94.6 mAP on the test set, outperforming the individual sensor networks by roughly 5%. We also described an implementation of the overall algorithm on a UAV platform and validated it in real-world experiments.

Full paper available here read more

PolyVerif: An Open-Source Environment for Autonomous Vehicle Validation and Verification Research Acceleration (2023)

Validation and Verification (V&V) of Artificial Intelligence (AI) based cyber physical systems such as Autonomous Vehicles (AVs) is currently a vexing and unsolved problem. AVs integrate subsystems in areas such as detection, sensor fusion, localization, perception, and path planning. 

Each of these subsystems contains significant AI content integrated with traditional hardware and software components. The complexity for validating even a subsystem is daunting and the task of validating the whole system is nearly impossible.

Full paper available here

Just another experiment done using the iseAuto autonomous shuttle (2022)

Real world experimentation is the key enabler for autonomous driving, test tracks are already full of vehicles driving autonomously, but to build effective technology, working in the real world, we need large experimentation campaings in any environment, here we have quite a challenging one! 


Practical path planning techniques in overtaking for autonomous shuttles (2022)

A reliable optimized sigmoid-based path planning algorithm that ensures smooth, fast and safe overtaking maneuver, while maintaining the necessary safety distance in shown in this research. In the proposed method, the desired smoothness of trajectories, the changes in steering angle and the lateral acceleration are controlled in a robust way. 

We describe the simulations, and the confirming real-world experiments, conducted using the autonomous shuttle iseAuto. Our results suggest that the sigmoid A-star algorithm leads to a smoother and more reliable motion when compared to other two standard methods. Specifically, the abruptness of necessary steering angle changes is reduced by factor of 4, and approaching the level of an experienced driver-like maneuver.

Full paper available here

Guest lecture for BIT (2021)

Building robots that perceive reality at least as good as humans do is considered a very complex task, and, with the help of artificial intelligence and sensor fusion, this goals is getting closer and closer.

In consideration of the most recent research developments in the field, this seminar aims to provide an insight and inspiration for graduate students on how the integration of sensory information from several sources can help robotic perception.

PDF presentation available here

Big data and data analysis (2021 - in Italiano) 

Repository creato per condividere slide e codici affrontati durante il corso IFOA2021 su big data e analisi dati. Il contenuto è reso disponibile e aperto, e copre argomenti che spaziano da aspetti generici riguardo ai big data, il trattamento di big data, e high performance computing.

Le slide descrivono i principali strumenti per il trattamento di big data, sistema di archiviazione basato su file system distribuito Hadoop, negoziatori delle risorse su cluster hadoop, YARN, spark, per raggiungere la principale fonte di utilizzo di big data, l'intelligenza artificiale nei suoi principi chiave.

Tutti i codici dei tutorial e le slide del corso sono disponibili sul repository github.

Dynamic fleet mission planning (2020)

This video illustrates our dynamic fleet mission planning method on topological maps, for autonomous underground mining operation. The video shows, at 20x the actual speed, the first hour of a four-hour shift in the mine shown in the figure below.

Vehicles that reach a terminal (either a loading site or an offloading site) can request re-optimization, once they have completed their stationary activity (loading or offloading). At that point, the GA-based optimizer generates a new mission for the vehicle in question, while also (potentially) modifying the missions of the other vehicles. Thus, the planning method must be able to handle the fact that vehicles move during optimization and must make sure to maintain causaility, by only making changes in those parts of the missions that lie beyond the end time of the optimization procedure.

Note also that the video has been speeded up by a factor 20. The actual traversal times are on the order of 7 - 15 minutes. For this rather small mine map, with a single off-loading station, 4-5 vehicles is close to the upper limit of what is reasonable. However, in bigger maps, the fleet mission optimizer can handle larger number of vehicles as well.

More information available on the Adaptive systems research group web page.

Development of a Serious Game to Enhance Assistive Rehabilitation (2018)


The aim of this study to investigate of novel assistive technologies based on serious gaming for the assessment of postural control and motor rehabilitation. Previous research already demonstrated that rehabilitation, assistive technologies, and physical activities can improve the quality of life of patients, and virtual reality applications may act as good additional companions during the therapeutic sessions. Indeed, workout routines supported by serious gaming encourage patients to train harder, bringing the therapy towards a pleasant game into a virtual environment, where specific goals must be reached. During the game, sensors track the movements of the players and transfer the data to software components that record a database from which patients’ progress can be determined. The acquired measures are interpreted in the form of new biomarkers, which enable the assessment of postural control. Such biomarkers are based on a probabilistic approach and show the capability to discriminate between well-performed exercises and incorrect movements. A long-term experimentation on a specific exercise is proposed, showing a considerable improvement, ranging from 5% to 30%, in the performance of patients .

The main contributions of this research are: i) the definition of new biomarkers for the postural assessment of patients affected by motor disorders; ii) the use of non-intrusive technologies, which enhances the freedom of movement for patients, increasing the reliability of the results; iii) the design of a virtual reality interface, which allows patients to interact in a pleasant and familiar environment without constant supervision; iv) the development of a new editor to easily customize virtual exercises, analyze rehabilitation progress, and create statistics.

Sohjoa Baltic - Bringing autonomous and eco-friendly public transportation into cities (2017)

Glad to be part of Sohjoa Baltic, an Interreg EU-funded project that aims to facilitate the transition to autonomous and eco-friendly public transport in the cities around the Baltic Sea. Sohjoa Baltic researches, promotes and pilots automated driverless electric minibuses as part of the public transport chain, especially for the first/last mile connectivity. The project brings knowledge and competence on organizing environmentally friendly and smart automated public transport. It also provides guidelines on legal and organizational setup needed for running such a service in an efficient way. Sohjoa Baltic projects brings autonomous small buses to drive demo routes in six Baltic Sea Region cities in the following years. The project involves 13 partners across 8 countries. Chalmers will, among other things, contribute with knowledge of vehicle engineering, autonomous technical development, intelligent cooperative driving behavior and risk analysis.

Labonatip 2.0 (2017)

The Lab-on-a-tip is a new open-source cross-platform software that enables facile configuration and use of the Fluicell BioPen system. The BioPen software enable independent control of each pressure line and solution delivery.

Lab-on-a-tip is based on QT5 library and released under the terms of the GNU GPL license.

See more on GitHub

Extension of Trajectory Planning in Parameterized Spaces to Articulated Vehicles (2017)

The main objective of this research is to study a novel method for safe maneuvering of articulated vehicles in warehouses. The presented method extends the concept of probabilistic planning on manifolds to articulated vehicles, which will be capable of driving, maneuvering and performing obstacle avoidance in any scenario. The proposed technique involves the extension of a parameterized space, developed for the reactive navigation of differential driven vehicles, to include an additional degree of freedom and use a probabilistic planner to calculate kinematically feasible trajectories. As a result, the algorithm is able to successfully generate maneuvers for an articulated truck and to navigate towards specific target points. The approach was validated using three problems representing different driving scenarios, demonstrating the possible utilization of the method in real-case scenarios. The solutions have been further benchmarked on multiple runs to evaluate success rate and to demonstrate the validity of the algorithm.


The Grand Cooperative Driving Challenge - GCDC2016

It works !! is good !!! The first time I heard this words in the truck I was so excited. Our autonomous driving truck is the result of months of hard work made by the entire Chalmers Truck team for the Grand Cooperative Driving Challenge GCDC 2016. What we achieved and what we all learned from this work has been incredible, thanks to the entire team, including who is not in this picture.

Team foto

Ants on the way ...

We are ants ! and we silently and stubbornly work to change the future!

ANTWaY - Automated Next generation Transport Vehicle for Work Yard application - project aims to solve the complex work yard transport problems.

This would involve developing systems for road vehicles, such as trucks or cars which can adapt to be completely autonomous while at a site using continuous guidance information from the site control system.

Challenges toward driverless technologies are on the way !

Simulation of a Laser based Driving Assistance for Smart Robotic Wheelchairs (2015)

This paper is presenting the ongoing work toward a novel driving assistance system of a robotic wheelchair, for people paralyzed from down the neck. The user’s head posture is tracked, to accordingly project a colored spot on the ground ahead, with a pan-tilt mounted laser. The laser dot on the ground represents a potential close range destination the operator wants to reach autonomously. The wheelchair is equipped with a low cost depth-camera (Kinect sensor) that models a traversability map in order to define if the designated destination is reachable or not by the chair. If reachable, the red laser dot turns green, and the operator can validate the wheelchair destination via an Electromyogram (EMG) device, detecting a specific group of muscle’s contraction. This validating action triggers the calculation of a path toward the laser pointed target, based on the traversability map. The wheelchair is then controlled to follow this path autonomously. In the future, the stream of 3D point cloud acquired during the process will be used to map and self localize the wheelchair in the environment, to be able to correct the estimate of the pose derived from the wheel’s encoders.

Demonstration of immersive reality technologies for design review (2014)

Along with CETMA - Area DIM there is a high active work on virtual reality and immersive systems

The amazing work of CETMA - Area Dim describing a novel hardware/software platform dedicated to design review process simplification, using immersive reality technologies. The proposed platform allows designers to interface with CAD engines and visualize different data types simultaneously into an immersive and stereoscopic multi-view visualization system. During the immersive sessions, the user can activate functionalities using cost-effective pointing devices and new conceived 3dUIs projected into the virtual environment. In this way, it is possible to easily manipulate virtual objects, perform basic operations such as rotations and translations but also more complex CAD functionalities such as surfaces shape modification. Each feature can be selected inside the virtual world using the smart 3D disk.

Users evaluations show that the use of a virtual environment may enhance the perception of designers ideas during the design process and the use of smart 3D interfaces simplifies the interaction among user and virtual objects.

CETMA link - Download [paper pdf], [poster];

New Traversability demo released (2014)

Try my new demo for traversability analysis in agricultural robotics: 

This application allows to load a single points cloud or a dataset, try filtering methods, frame reference transformations and traversability analysis.

Follow using instructions here.

Road surface analysis for driving assisntace  (2013)

In order to increase the level of driving automation in future cars, it is important to address critical issues, including road monitoring for irregularities and damage detection.

The primary scientific aim of this research is to investigate the problem of road surface analysis in urban and extra-urban scenarios for driving assistance purposes towards the final goal of implement such technologies on future driverless cars. The proposed approach uses a range sensor to generate an environment representation in terms of 3D point cloud that is then processed by a normal vector-based analysis. Even small irregularities of the road surface can be successfully detected, using such information to warn the driver or enable an autonomous vehicle to regulate its speed and change its course appropriately.

Download the full paper and presentation of the 3DRP-PCL2014

RRT path planning in TP-Space for hybrid navigation  (2013)

This work addresses hybrid reactive-planned navigation for autonomous vehicles with non-holonomic constraints. Hybrid methods have the potential to combine the strengths of reactive methods, e.g. fast response to dynamic or poorly mapped environments, while avoiding their main pitfalls: (i) the possibility of getting stuck in a local minimum and (ii) not being aware of global path optimality. In order to achieve this objective, we propose extending Rapidly-exploring Random Tree (RRT) planners to Trajectory Parameter Space (TP-Space), previously proposed as an efficient approach to detect collision-free paths of any-shape, kinematically-constrained vehicles. As a result, our proposal generates a tree whose edges are all kinematically-feasible paths, which can be followed by a reactive navigation engine. Our initial experiments demonstrate the suitability of such a hybrid navigator for real time operation with a simulated Ackerman-steering vehicle. Moreover, it is shown how simultaneously employing several families of trajectories to expand the tree improves the obtained plans.


Kinect-based parking assistance system  (2012)

This work presents an IR-based system for parking assistance and obstacle detection in the automotive field that employs the Microsoft Kinect camera for fast 3D point cloud reconstruction. In contrast to previous research that attempts to explicitly identify obstacles, the proposed system aims to detect “reachable regions” of the environment, i.e., those regions where the vehicle can drive to from its current position. A user-friendly 2D traversability grid of cells is generated and used as a visual aid for parking assistance. Given a raw 3D point cloud, first each point is mapped into individual cells, then, the elevation information is used within a graph-based algorithm to label a given cell as traversable or non-traversable. Following this rationale, positive and negative obstacles, as well as unknown regions can be implicitly detected. Additionally, no flat-world assumption is required. Experimental results, obtained from the system in typical parking scenarios, are presented showing its effectiveness for scene interpretation and detection of several types of obstacle.

Download the full paper and datasets.

Vision in agricultural environment  (2012)

This research aims to address the issue of safe navigation for autonomous vehicles in highly challenging outdoor environments. Indeed, robust navigation of autonomous mobile robots over long distances requires advanced perception means for terrain traversability assessment. The use of visual systems may represent an efficient solution. This paper discusses recent findings in terrain traversability analysis from RGB-D images. In this context, the concept of point as described only by its Cartesian coordinates is reinterpreted in terms of local description. As a result, a novel descriptor for inferring the traversability of a terrain through its 3D representation, referred to as the unevenness point descriptor (UPD), is conceived. This descriptor features robustness and simplicity. The UPD-based algorithm shows robust terrain perception capabilities in both indoor and outdoor environment. The algorithm is able to detect obstacles and terrain irregularities. The system performance is validated in field experiments in both indoor and outdoor environments. The UPD enhances the interpretation of 3D scene to improve the ambient awareness of unmanned vehicles. The larger implications of this method reside in its applicability for path planning purposes. This paper describes a visual algorithm for traversability assessment based on normal vectors analysis. The algorithm is simple and efficient providing fast real-time implementation, since the UPD does not require any data processing or previously generated digital elevation map to classify the scene. Moreover, it defines a local descriptor, which can be of general value for segmentation purposes of 3D point clouds and allows the underlining geometric pattern associated with each single 3D point to be fully captured and difficult scenarios to be correctly handled.

Download the full paper