Cognitive dynamic systems Lab Center for Cognitive Modeling MIPT
Cognitive Dynamic Systems Laboratory specializes in the development of a strategic level of behavioral control of a cognitive agent or robotic platform. To maintain autonomous complex adaptive behavior, it is necessary to implement such functions as the acquisition, presentation and use of knowledge, modeling reasoning, generation of a behavior plan, distribution of roles, goal setting and reflection.
The laboratory employs an interdisciplinary team that uses modern ideas about the work of the psyche and the functioning of the human brain to build new algorithms for machine learning with reinforcement, new architectures of neural networks, new methods for representing knowledge and planning behavior. The laboratory develops original fundamental approaches and creates applied intelligent systems on their basis.
Head of the laboratory cognitive dynamic systems
For solving complex problems in the field of artificial intelligence, it is important to have excellent mathematical training in discrete mathematics, statistics and confident knowledge of working with programming languages and basic Python \ C ++ libraries. In our laboratory, students and graduate students work in a friendly team, learn to apply their knowledge in real scientific projects, get new world-class results, participate in leading conferences with their results, publish articles in journals on machine learning, computer vision and others. We set ourselves ambitious goals and motivated guys can always find an interesting task for themselves.
Each group of the Center works in three directions: 1. Support of the corresponding subsystem in the general system of the cognitive agent controlled by the STRL architecture, in which, most often, standard debugged algorithms are used that effectively solve the necessary subtasks (for example, for a "tactical" group, this is an algorithm mapping from lidar data, which can be based on a standard method built into ROS).
2. Development of fundamental specific, often narrowly specialized, original methods for solving subtasks of the group (for example, a new method of organizing schemes for reinforcement learning based on a model, a new architecture for detecting traffic signals for the Jetson architecture, new algorithms for the propagation of activity in a semiotic network, etc.).
The results obtained are used for the publication of articles in high-ranking conferences and journals, for the defense of master's and master's theses.
3. Participation in short-term competitions both on narrow topics of the group (for example, MineRL and VQA for a "strategic" group), and in a general robotics competition, which demonstrates the general operation of the system being developed (for example, EuroBot).
The main practical result of the Center's work is a modular software control system similar to the Robotic Operation System (ROS), which implements the main STRL subsystems and is built on the principles of partially open source code.
Re-identification of objects, tracking and trajectory prediction
Identification, tracking and prediction of the trajectory of movement of dynamic objects based on the integration of data received from various on-board vision systems of a mobile robot
We are always glad to see motivated students and graduate students in our laboratory. One of the options to join the projects of the Center for Cognitive Modeling is the opportunity to apply for special targeted places provided by sponsors for the master's or postgraduate studies at MIPT.