I am a Ph.D. candidate at the department of Computer Science at Brown University. I am advised by Professor Iris Bahar. Previously, I was at Drexel University where I earned an integrated BS and MS in Electrical Engineering, and a minor in Computer Science.

My research interest lies in embedding robots with a causal understanding of the world to facilitate smarter decision making under uncertainty. Human decision making is guided by causal-reasoning. Transferring human causal knowledge to robots can pave the way for generalizable autonomous planning in unknown environments.

In my free time, I like to time-travel (aka read fiction), scout for cafes, explore New England small towns and live the slow life.

Semanti Basu

• We developed a model to embed human-generated causal reasoning into Partially Observable Markov Decision Processes (POMDPs) to allow robots to plan better under uncertainty. We built an interface to allow seamless transfer of human causal understanding to robots in the form of causal graphs. We demonstrated a 2X improvement in robot planning performance in object assembly when it is causally informed. We are exploring several directions on how to aid autonomous decision making through human-generated causal models.

• Designed and prototyped a stereo-depth estimation algorithm to strike a balance between compute time, compute power and accuracy. Recent developments in stereo depth estimation show great performance on optics-limited scenes (ex: transparent/ translucent objects, texture less regions etc). However, they are often much more time consuming and resource intensive than traditional methods. Developed a hybrid approach to leverage the strengths of traditional optimization with the power of learning to improve performance by 2X

• Depth Estimation in time and resource constrained environments (Amazon Science internship)

• Publications

• Developed an architecture of Convolutional neural networks to predict anthropometric points on human silhouettes. Developed a biometric-reidentification method (96% accuracy) that leveraged the points predicted for 3D human body classification.

• Salient point detection and 3D human body classification

• Problems that require planning under uncertainty are often modeled as Partially Observable Markov Decision Processes (POMDPs). Its intractable to solve POMDPs exactly, so Monte Carlo sampling based methods are often used. These become increasingly slow with increased domain size. We are looked into several ways to accelerate planning and achieved upto 2X improvement over serial POMCP.

• Performance aware POMDP planning

Semanti Basu, Peter Bajcsy, Thomas Cleveland, Manuel Carrasco, R. Bahar

• Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies - Volume 2: BIOIMAGING, 2023

• LipoPose: Adapting Cellpose to Lipid Nanoparticle Segmentation

Semanti Basu, Chenxi Li, Fernand Cohen

• Multimedia Tools and Applications, 2022

• Anthropometric salient points and convolutional neural network (CNN) for 3D human body classification

• Semir Tatlidil*, Semanti Basu*, Kaishuo Zhang, F. Tao Burga Montoya, R. Iris Bahar, Steven Sloman.In Cognitive Modeling in Robot Learning for Adaptive Human-Robot Interactions Workshop @ ICRA 2023.

• Object assembly using inferred causal models from humans.

Semanti Basu

• Undergraduate thesis, Drexel University, 2020

• A Convolutional Neural Network-based Approach to Personalized 3D Modeling of the Human Body and Its Classification

Semanti Basu, Sreshtaa Rajesh, Kaiyu Zheng, Stefanie Tellex, R Iris Bahar

• Rss workshop on software tools for real-time optimal control, 2021

• Parallelizing POMCP to solve complex POMDPs

Semanti Basu, Semir Tatlidil, Kaishuo Zhang, F. Tao Burga Montoya, R. Iris Bahar, Steven Sloman

• Workshop YOUR Study Design!, A Workshop at the ACM/IEEE International Conference on Human-Robot Interaction, 2023

• Human-Generated Causal Models to Aid Robot Planning