IEEE Kharagpur Section

Self-navigation in crowded environments poses a significant challenge for multi-agent systems with non-holonomic robots relying on local sensing. The talk covers a fast, sensor-driven navigation controller that computes control commands for safe maneuvering among non-cooperating agents. Our approach introduces an input-constrained feedback controller specifically designed for non-holonomic robots, with a focus on invariant sets to ensure stability and safety. These invariant sets guide the robots toward their target while allowing for direct computation of safe control inputs, eliminating the need for pre-planned paths. Speaker(s): Leena Vachhani, Virtual: https://events.vtools.ieee.org/m/441459

Interpreting visual signals introduces both challenges and opportunities in the design of control and autonomous systems. This talk will explore two key concepts that address these challenges. In the first part, I will introduceperception contracts—an innovative approach to analyzing visual control systems that rely on Deep Neural Networks for state estimation. A perception contract provides an over-approximation of a state estimator while guaranteeing closed-loop system invariants. These contracts can be automatically synthesized using data and model-based analysis and have been successfully applied to systems such as automated landing controllers and lane-keeping systems. The second part of the talk will focus on algorithms for computing indistinguishable sets—sets of states that cannot be distinguished based on available visual data. These sets help define the theoretical limits of visual control, revealing the boundaries of what can be achieved with coarse measurements in dynamic environments. Throughout the talk, I will mention various examples, highlight the tools available, and discuss open problems that invite further exploration in this area. Speaker(s): Sayan Mitra, Virtual: https://events.vtools.ieee.org/m/441460

Retinal detachment (RD) is a severe disorder that leads to vision loss, although it can be highly treatable with prompt and appropriate medical intervention. Early detection of RD can increase the probability of successful reattachment and improve visual outcomes, mainly before the macular involvement. Manual screening of RD is tedious and laborious, making it challenging to implement on a wide scale of healthcare applications. Therefore, an automated screening tool for early RD detection is very essential. This work proposes a novel multiclass RD grading framework using a Graph neural network (GNN) model. For this study, the simple linear iterative clustering method has been employed, which considers all samples, training, and testing samples as nodes and establishes a set of edges or connections between them to form a graph structure. In addition, three graph neural networks, namely the graph convolutional network, GraphSAGE, and graph attention network, were used as feature extractors to learn the graph's semantic relations and local-global features effectively. Finally, an ensemble learning approach with a majority voting mechanism was utilized to assign weights to the retrieved graph features, leading to the final prediction. Speaker(s): Dr. R. Murugan, Virtual: https://events.vtools.ieee.org/m/442009

Funnel control is a powerful and simple nonlinear control method to solve the tracking problem for uncertain systems with prescribed transient performance as well as with guaranteed accuracy. The origin of the funnel controller can be seen in (adaptive) high gain feedback and the funnel controller resolves the major problems of high-gain feedback; in particular, arbitrary reference signals can be tracked arbitrary well and the feedback gain is not monotonically increasing. A simple proof is provided why funnel controller works and further theoretical extensions are discussed and illustrated with experimental setups and simulations. Speaker(s): Stephan Trenn, Virtual: https://events.vtools.ieee.org/m/441462