We explored a method to accurately (we achieve >81% accuracy) locate and resolve ambiguous portions of instructions related to common household tasks given to robots.

The Human Comfort Classifier (HCC) detects discomfort in real time from video, combining pose estimation, facial landmarks, and sentiment analysis in a transparent, rule-based model. This approach is beneficial for social skills training and achieves 78% accuracy without relying on deep learning, ensuring explainable feedback for users.

BB-Align is a two-stage framework that improves pose accuracy in vehicle-to-vehicle (V2V) perception by using Lidar Bird’s-eye View (BV) images and bounding boxes. It aligns data from different cars, correcting pose errors without requiring extra training, making it suitable for real-world V2V applications. Tests show BB-Align enhances pose accuracy, achieving errors under 1m and 1°, and improves object detection in error-prone scenarios.

This research focused on using AutoDock Vina to model protein-ligand binding for kinase inhibitors like Regorafenib, targeting Alzheimer’s pathology by inhibiting tau hyperphosphorylation and amyloid-beta aggregation. This approach aims to identify effective compounds that can slow disease progression.

The project tests a centralized control system using computer vision to manage vehicle and pedestrian flow at intersections. By employing YOLOv8 for real-time detection, it automates queuing to reduce wait times and improve safety.