C-BRIC Researchers at Purdue and UPenn Slated to Present at ICRA

C-BRIC researchers will present their work at the 2023 International Conference on Robotics and Automation (ICRA). The IEEE-sponsored conference will be held from May 29-June 2, 2023, in London.

 

Purdue researchers Adarsh Kosta, Chamika Liyanagedera, Manish Nagaraj, Sourav Sanyal, and Kaushik Roy and University of Pennsylvania researchers Fernando Cladera, Beiming Li, Xu Liu, Ian D. Miller, Ankit Prabhu, Yuezhan Tao, Dinesh Thakur, Yuwei Wu, Alex Zhou, Lifeng Zhou, and Vijay Kumar will present their C-BRIC work at ICRA.

 

The Purdue team of Adarsh Kosta and Kaushik Roy will present their work on "Adaptive-SpikeNet: Event-based Optical Flow Estimation using Spiking Neural Networks with Learnable Neuronal Dynamics" at ICRA. In this work, they proposed an "adaptive fully-spiking framework with learnable neuronal dynamics to alleviate spike vanishing problems. [They use] surrogate gradient-based backpropagation through time (BPTT) to train [their] deep SNNs from scratch. [They validated] their approach for the task of optical flow estimation on a Multi-Vehicle Stereo Event-Camera (MVSEC) dataset and a DSEC-Flow dataset. [Their SNNs] offer substantial savings in network parameters and computational energy while attaining lower Edge Placement Error." Kosta is currently a PhD student under the direction of Kaushik Roy. Roy is a faculty member at Purdue's Elmore Family School of Electrical and Computer Engineering and the Director of C-BRIC.

 

Purdue's Chamika Liyanagedera, Manish Nagaraj, and Kaushik Roy will present "DOTIE - Detecting Objects through Temporal Isolation of Events" at ICRA. In this paper, the group proposed "a novel technique that utilizes the temporal information inherently present in events to efficiently detect moving objects. [The] technique consists of a lightweight spiking neural architecture that is able to separate events based on the speed of the corresponding objects. These events are further grouped spatially to determine object boundaries. This method of object detection is both asynchronous and robust to camera noise. By [using this] architecture, autonomous navigation systems can have minimal latency and energy overheads for performing object detection." Liyanagedera is a postdoctoral researcher and Nagaraj is a PhD student, both under the direction of Kaushik Roy. Roy is a faculty member at Purdue's Elmore Family School of Electrical and Computer Engineering and the Director of C-BRIC.

 

Sourav Sanyal and Kaushik Roy, C-BRIC researchers at Purdue University, published their work on "RAMP-Net: A Robust Adaptive MPC for Quadrotors via Physics-informed Neural Network" at the conference. This publication focused on a "Robust Adaptive MPC framework via PINNs (RAMP-Net). The method "uses a neural network trained partly from simple [ordinary differential equations] and partly from data. [The] analytical functions inside the loss function [enforces] robust behavior for parametric uncertainties [while] a regular data loss is used for adapting to residual disturbances. [They reported] reduction in tracking errors for speeds ranging from 0.5 to 1.75 m/s compared to two [state-of-the-art] regression based [model predictive control] methods." Sanyal is a PhD student in Kaushik Roy's research group. Roy is a faculty member at Purdue's Elmore Family School of Electrical and Computer Engineering and the Director of C-BRIC.

 

C-BRIC researchers from the University of Pennsylvania, Fernando Cladera, Xu Liu, Ian D. Miller, Ankit Prabhu, Lifeng Zhou, and Vijay Kumar, along with collaborator Camillo J. Taylor will present "Active Metric-Semantic Mapping by Multiple Aerial Robots." This work focuses on their approach to the "active metric-semantic mapping problem that enables multiple heterogeneous robots to collaboratively build a map of the environment. The robots actively explore to minimize the uncertainties in both semantic (object classification) and geometric (object modeling) information. The environment [is represented by] using informative but sparse object models. [The model is used to] select actions for each robot to minimize uncertainties. The proposed framework is applicable to a wide range of realworld problems." Cladera, Liu, and Miller are PhD students under the direction of Vijay Kumar. Prabhu is an MS student and Zhou is a former postdoctoral researcher, both with Kumar. Kumar is a faculty member and dean in the School of Engineering and Applied Science at the University of Pennsylvania. Taylor is a faculty member in the Computer and Information Science Department at the University of Pennsylvania.