Research Interests

My primary research interests are in:

  • Exploring the interplay between learning and control, and understanding how combining the two can lead to an improved control performance.

  • Efficient computation of reachable sets.

  • Applications of reachability theory in multi-agent autonomous systems, like Unmanned Aerial Vehicles (UAV) Traffic Management, etc.

Selected Projects

Robust Sequential Path Planning


Recently, there has been immense surge of interest in using unmanned aerial vehicles (UAVs) for civilian applications in projects like Amazon Prime Air and Google Project Wing. These applications include package delivery, fire-fighting, aerial surveillance, and fast disaster response. As a result, UAV traffic management (UTM) systems are needed to support potentially thousands of UAVs flying simultaneously in the airspace, in order to ensure that their liveness and safety requirements are met.

One essential problem that needs to be addressed is how a group of vehicles in the same vicinity can reach their destinations while avoiding collision with each other. Hamilton-Jacobi (HJ) reachability is ideal for analyzing such safety-critical problems because it provides safety guarantees and is flexible in terms of system dynamics; however, due to the curse of dimensionality, the analysis of simultaneously many vehicles is intractable without imposing additional assumptions or structure. One such proposed structure is to assign priorities to vehicles; path-planning problem is then solved sequentially starting from the highest priority vehicle. In this project, we focus on designing the sequential path-planning algorithms that are robust to disturbances and intruders, and thus guarantee collision avoidance.

(Image credits: NASA)

Relevant publications:

  • M. Chen, S. Bansal, J.F. Fisac and C.J. Tomlin, “Robust Sequential Path Planning Under Disturbances and Adversarial Intruder,” IEEE Transactions on Control Systems Technology (submitted) (PDF)

  • S. Bansal*, M. Chen*, J.F. Fisac and C.J. Tomlin, “Safe Sequential Path Planning of Multi-Vehicle Systems Under Disturbances and Imperfect Information,” IEEE American Control Conference, May 2017 (PDF)

Learning-based system identification/optimization for improved control


System identification, the mathematical modeling of a system’s dynamics, is one of the most basic and important components of control. Constructing an appropriate model is often the first step in designing a controller. Depending on the system, however, it might be extremely challenging, if not impossible, to come up with a system model starting from the first principles. To circumvent these modeling issues, data-driven, learning-based approaches provide a promising alternative. In this work, we explore different machines learning tools for system identification purposes, with an emphasis on developing architectures to efficiently combine them with the control theory tools.

Relevant publications:

  • S. Bansal, R. Calandra, T. Xiao, S. Levine and C.J. Tomlin, “Goal-Driven Dynamics Learning via Bayesian Optimization,” 56th IEEE Conference on Decision and Control (submitted) (PDF)

  • S. Bansal*, A.K. Akametalu*, F.J. Jiang, F. Laine and C.J. Tomlin, “Learning Quadrotor Dynamics Using Neural Network for Flight Control,” 55th IEEE Conference on Decision and Control, Dec 2016 (PDF, Video)

Model Predictive Control (MPC) for load shaping and voltage control in smart grids


Electricity power systems which traditionally generate electricity energy from fossil fuels and transmit it to demand are moving towards a new stage. In order to mitigate environmental pollution and climate change, sustainable/renewable energy resources have been largely integrated into power grids. However, since the renewables are highly intermittent and usually cannot match the load, the local energy balancing becomes a big challenge; this unbalance between supply and demand can cause a significant rise in the voltage level in the grid. Moreover, over-generation during off-peak hours can cause load leveling problems and may result in so called “duck curve.”

The good news is that flexible loads (like Electric vehicles, batteries, etc.), even though represent an additional load on the distribution systems, are controllable and thereby offers an important opportunity. In this work, we explore how flexible loads can be used to flatten the load profile during the period of over-generation (also referred to as the “belly” of duck).

(Image credits: Electric Mobility SW)

Relevant publications:

  • C.L. Floch, S. Bansal, C.J. Tomlin, S. Moura and M. Zeilinger, “Plug-and-Play Model Predictive Control for Load Shaping and Voltage Control in Smart Grids,” IEEE Transactions on Smart Grid, Jan. 2017 (PDF)

  • S. Bansal, M. Zeilinger and C.J. Tomlin, “Plug-and-Play Model Predictive Control for Electrical Vehicle Charging and Voltage Control in Smart Grids,” 53rd IEEE Conference on Decision and Control, Dec 2014 (PDF)