ARTIFICIAL INTELLIGENCE
in dynamiс action
Dynamic control and decision support powered by AI
Reinforcement learning
AI applications in: robotics, computer vision, algotrading, automation etc.
Projects
Advanced driver assistance systems
● Developed a neural-network based solution for lane departure warning and virtual corridor system
● Developed a neural-network based solution for lane departure warning and virtual corridor system
● Developed a neural-network based solution for lane departure warning and virtual corridor system
● Implemented and tested the solution on the vehicle’s end ECU
Research
Predictive reinforcement learning
- Designed a novel framework of generalized predictive controllers
- Designed stabilizing constraints
- Combined value-iteration and Q-learning agents with predictive controllers (called stacked RL)
- Demonstrated performance improvement potentials
- New agent is safe by design
Stabilizing reinforcement learning
- Designed a novel framework for reinforcement learning with online stability guarantee
- Environment (system) stability is ensured in every learning episode
- Experimentally verified
Stable locomotion of legged robots
- Implemented and tested a footstep planner
- Implemented odometry based on foot contacts
- Implemented and tested a nonlinear model-predictive controller for reaction forces
- Developed hardware interface for QuadDSK
- Prototyped policy gradient (PPO) agent to learn to gallop
Field swarm robot with ground condition identification and optimal propulsion control functionality
- German patent
- Invention highlighted in a number of publications (see, for instance, this)
Fundamental research in control theory
- A number of theorems and methods developed
- Strong focus on explicit account of real-world controller implementation
- Developed a framework for control system analysis and synthesis under computational uncertainty based on constructive mathematics (video)
- Developed methods of most general stochastic stabilization of nonlinear dynamical systems under digital controllers
- Widely employed nonsmooth analysis and stochastic process theory
- Proved some constructive theorems for optimal control
- Proved a constructive theorem on stabilization via sliding-mode control and did an explicit computation of controller sampling time in a practical example of traction control
Courses
Reinforcement learning
- Tabular methods
- Policy gradient
- Actor-critic
- Convergence
- Deep reinforcement learning
- Predictive reinforcement learning
- Stability & safety
Advanced control methods
- Stability theory of nonlinear dynamical systems
- Lyapunov-based controller and observer design
- Model-predictive control
- Sliding-mode control
- Adaptive, robust, fault-tolerant control
- Sample-and-hold control
- Watch introductory video
Miscellaneous courses
- Safety and robustness aspects of artificial intelligence
- Mathematical modeling and system identification
- Complex systems
- Systems theory
- Optimal control
- Vehicle control
- MATLAB/Simulink
Team
Pavel Osinenko (group head)
- Expert in control systems and reinforcement learning
- 40+ Scopus‑indexed publications incl. 5 CORE A conferences and about 20 in Q1 journals
- Many years of teaching experience at the lecturer level in English and German
- Experience in many research projects in Germany, Russia as PI and lead engineer
12 PhD students/researchers + 5 in Germany
8 Msc students
4 Engineers
Alumni: over 20 MSc, 1 PhD
Partners
