Selected industrial projects
I-nav | Anomaly Detection | Bin-Picking | B3P | Kerberos | past
I-nav - industrial navigation framework for transport robots
I-nav is a software framework for mobile robot navigation in industrial environment, primarily targeted to material transportation between fixed pre-defined places.
Main advantage of this system in comparison to classical approaches is full autonomy without need of costly and fixed installations for navigation purposes. Since the autonomous navigation is based on localization, which uses natural features of the environment, it is relatively fast and easy to adjust current paths and destinations or add new.
Modular structure of the system allows feasible extensibility according to customer needs.
Basic version of the system includes:
- navigation along fixed paths, which need to be learned by manual drive between defined places,
- chassis control for 4-driven wheel kinematics.
Anomaly detection and classification
Modular software toolkit for visual navigation, anomaly detection, and classification in autonomous manipulation. Approaches used in computer vision are often sensitive to appearance changes and suffer from uncertainties. We have developed a powerful anomaly detection and classification toolkit based on bayesian neural networks. The developed method accounts for the uncertainties in real data and provides additional classification results in the form of a probability distribution. Apart from traditional classification results, the output of our method can be used for in-depth analysis of the data and its underlying distribution and similarities. The developed methods were integrated into a complex planning and navigation system for autonomous robotic manipulation.
Software solutions for smart bin-picking
Software toolkit of developed methods and solutions for smart bin-picking based on the use of 3D or 2D vision systems for object detection and localization, coupled with a robotic manipulator. Approaches combining traditional machine vision techniques with deep learning methods were applied and tested on various laboratory demonstrators, resulting in a set of software tools, which can be readily adjusted depending on the application. The flexibility of the investigated solutions proves to be an advantage over currently available off-the-shelf part picking solutions. (Continuation of the work on the B3P feasibility study.)
B3P – Škoda Auto, Random Bin-Picking
Project runtime: Oct 2016 to Mar 2017
Project lead: Libor Preucil (CTU/CIIRC), in cooperation with CTU/FEE (Pavel Burget)
Feasibility study, design dvelopment and rapid prototyping of laboratory solution and reliability tests with the aim to aply the solution in a real production line. The project addresses automated robotic picking of partly organized parts (steel pots) from a storage bin with the goal to feed spark welding assembly cells (car chassis manufacturing).
Kerberos - security car scanner
Laser-based under-carriage vehicle inspection system developed in cooperation with VOP CZ sro. The system is able to recover a full three-dimensional relief of a passing-over vehicle, compare it with a save vehicle scan from a database and identify any foreign objects. Unlike other inspection systems, which use high-resolution cameras, the Kerberos system employs laser rangefinders and thus allows to find objects with the same texture as the component they are attached to. Its basic working principle is based on two- and three-dimensional shape reconstruction and registration methods, commonly used for mobile robot localization and mapping. The cooperation of the CTU’s Intelligent and Mobile Robotics Lab with the VOP, inc. allowed a swift transfer of the aforementioned state-of the art methods to the security domain. The vehicle inspection scanner is currently installed at two heavily safeguarded facilities in Czech Republic.
Ven ze tmy
The aim of the project was to develop a new kind of an navigation aid for visually impaired. The project was managed by Integrace. We worked on the obstacle detection system, which was based on a stereo camera.
The aid consists of the sensor for obstacle detection and a tactile display which is integrated to a shirt. The tactile display has 7x3 vibrating actuators. The position of an obstacle in front of the user determines which actuator is activated. This allows the user to estimate positions of the surrounding obstacles.
Microsoft CZ: .NET Micro Framework for real robotic applications
The aim of the project is to undertake case studies and user evaluation of the properties of MS NET Micro Framework in relation to its practical applications in management and decision making in autonomous intelligent systems (intelligent robotics) and real-time utilization. The project focuses on the initial analysis of the problems with defining the usual requirements imposed on. The project examines and verifies the admissibility and availability of solutions to fundamental problems of mobile robot reactive navigation through the NET Micro Framework. Thereafter it provides a quantitative and qualitative evaluation of the properties of the system and formulates the resulting limits of applicability and recommendations for users. The system experimental verification has been performed on an intelligent wheelchair platform.
About:solution s.r.o.: Planning for vehicle transportation in Czech Republic.
Transportation or distribution problems are practical issues, which often belongs to NP or even NP-hard class type of problems. Therefore approaches developed in Artificial Intelligence can be useful in many practical situations, as they can provide sufficiently good solution within a fraction of time required to find optimal solution due to complexity of the problem. This is also the case of vehicle planning in emergency situations in which a set of locations have to be visited by a particular vehicle under given time limit. We exploited are knowledge from the multi-goal path planning for mobile robots in this type of transportation problem that is formulated as an extended version of the well-known multiple traveling salesman problem with min-max criterion. The extension includes: 1) multiple starting cites (multi-depot); 2) initial penalization of salesmen; 3) non zero costs of cities; 4) possible specification of closed/opened paths for particular salesmen. Besides, we also addressed the problem of minimal number of required salesmen (vehicles) that are needed to fulfill the mission while the solution quality satisfies the given time limit.
Upon request of the about:solution s.r.o. company we developed a solver addressing the generalized variant of the MTSP-MinMax. The solver is based on the self-organizing map that provides real-time response required by operators, while found solutions meet required properties. We provide a solution of planning problem for vehicle transportation in Czech Republic.
Safety Critical Systems Design
The Safety Critical Systems need specific methods for design, development and testing. The Gerstner laboratory takes part in development of the railway Electronic Automatic Line Block system produced by AŽD Ltd. company. This system is responsible for control of optical signals and signals sent to train and for position detection of the train.
The critical part of the system, the functional safety, is safeguarded by redundant means mainly at the control level. The formal specification of selected critical functions of the system were used to improve the safety of the whole system. This system was awarded by a golden medal at the Czech Engineering Fair, Brno, in 2001.
CyberSonic System
The Gerstner Laboratory developed a sonar measuring system for Povodi Labe company. Main task of this system is to make a 3D map of a river bed and to visualize river bed profile in order to locate obstacles in the fairway and keep river Labe navigable.
The special measuring ship equipment consists of a sonar arm with 12 sonar transducers and two GPS receivers for localization. The system relies on a kinematic model of the ship to improve precision of the localization. The measured sonar data are fused by grid-based methods. The system is able to process maps in WGS84 or JTSK geodetic system.
Train Locator
Modern railway control and safety systems need to be aware of real- time localization of trains. The Gerstner Laboratory participated at a project solving the railway vehicle localization problem for the cases, where global positioning information (like GPS) is temporarily unavailable (tunnels, deep valleys, etc.).
The given solution also assumes no additional landmarks or her extraordinary installations aside the train track. The realized approach is based on smart fusion of onboard- gathered data from vehicle odometer and accelerometer. GPS signal is used as a calibration tool for developed algorithms.
Run-time Diagnostic Systems
Diagnostic systems are necessary to maintain and control wide network of safety and technology systems. The concept and SW for Local Diagnostic System (LDS) for Czech Railways were designed and developed. LDS represents a modular diagnostic system for collection, storing, evaluation and monitoring of run-time status data of the control and safety technology.
Diagnostic Local Server collects, filters and stores data for offline retrieval as well as generates diagnostic reports based on data analysis and enables data access to intra-net. The system also provides service messages and notices to the maintenance staff through a GSM module. The LDS upper layer is the Global Diagnostic System (GDS) enabling data transmission via data network from individual LDS clusters to location of a centralized supervision.
See also our research and finished projects.