Abstract: Autonomous search for specific objects in an unknown environment is a common task across a wide range of assistive and service applications, including human-robot assistance, warehouse operations and object retrieval. Robots engaged in such operations must perform the tasks in practical scenarios that include occluded environments and/or limited illumination. Existing methods exhibit good performance under ideal conditions, but their performance drastically drops under practical scenarios due to limited and degraded visibility of the target object. Also, their high computational requirements limit their deployment on embedded platforms with constraints on the resources. To address these challenges, this work proposes a lightweight framework for autonomous search under low-light and occlusion conditions. It is designed to handle in particular mobility aids. The proposed framework is a two-stage transformer-based network that is guided by low-light enhancement and occlusion-aware loss functions. The proposed model is evaluated under various illumination settings and various occlusion levels. Experimental results with a wheeled robot searching for multiple mobility aids in an unknown indoor environment are presented.

Event Details
Title: A lightweight model for autonomous mobility aid search amidst occlusions and low light
Date: September 28, 2026 at 3:00 PM
Venue: ESB 246C
Speaker: Mr. Nambala Ramsai (EE20D056)
Guide: Dr. Sridharan K
Type: PHD seminar

Updated: