IIIT-H robotics research moves from lab to factory floor

HYDERABAD: Research by the Robotics Research Center (RRC) at the International Institute of Information Technology, Hyderabad (IIIT-H), is moving towards real-world deployment through a three-year collaboration with Ati Robotics, with the institute’s vision-based navigation technology being tested in an industrial setting. In a recent demonstration at Ati Robotics’ factory, an autonomous mobile robot (AMR) navigated […]HYDERABAD: Research by the Robotics Research Center (RRC) at the International Institute of Information Technology, Hyderabad (IIIT-H), is moving towards real-world deployment through a three-year collaboration with Ati Robotics, with the institute’s vision-based navigation technology being tested in an industrial setting. In a recent demonstration at Ati Robotics’ factory, an autonomous mobile robot (AMR) navigated to a specified object using a camera, images captured during an earlier walkthrough and an artificial intelligence model that reconstructed three-dimensional geometry from image pairs. The demonstration stood out because the robot did not rely on a conventional LiDAR occupancy grid, scan-matched position estimates or a metric map. The work explores whether robots can navigate using visual cues and recognisable landmarks, much as humans use their surroundings to identify locations and choose routes. Researchers explore camera-based navigation for robots LiDAR, or light detection and ranging, is widely used by autonomous mobile robots to map surroundings and estimate their position. It often works alongside simultaneous localisation and mapping (SLAM), a technique that allows a robot to build a map while tracking its location within it. The IIIT-H team has been exploring alternatives that use visual information and sparse representations of the environment, potentially reducing the storage and computational requirements associated with dense maps. Explaining the approach, Prof. Madhava Krishna said humans generally navigate by recognising surroundings, remembering visual cues and adjusting their routes as conditions change, rather than relying on precise three-dimensional coordinates. In sparse localisation, a robot estimates its position using a limited set of recognisable landmarks instead of representing every part of its surroundings in detail. Outdoors, these could include buildings, traffic signals, trees and shops. In warehouses, landmarks could include a doorway, an elevator or a particular section of a storage area. Image-based navigation uses visual cues to reach destinations The team also investigated image-based navigation, in which a robot is directed towards a point identified in an image rather than through a set of precise coordinates. In a warehouse, the target could be a pallet or a docking station. The system compares images of the same area to identify corresponding pixels, estimates their three-dimensional relationships and generates a cost image to assess which directions are more suitable for movement. This approach allows the robot to use visual information to guide its movement without depending entirely on a conventional three-dimensional coordinate map. Three-year collaboration translates academic research into industry Prof. Sourav Garg of the Robotics Research Center said the research began in 2023 with a question about how humans navigate visually using semantics, topology and local geometry. "We built a navigation stack that defines mapping, localization, planning and control simply through relative connectivity of image segments," he said. Garg said the navigation system’s modular design allowed the industry partner to retain or replace components according to its requirements. Ati Robotics subsequently developed OccupancyReact, a LiDAR-based variant that is closer to the company’s existing deployed solutions. Ati Robotics focuses on precision during final approach Ati Robotics said it already had vision-driven navigation running on its hardware, while continuing to work on what it described as Physical AI. The company said its current focus was on the final two metres of a robot’s journey, when it docks, approaches a pallet or manoeuvres through a narrow bay. Improvements at this stage could save time across repeated operations during a work shift, the company said. The collaboration illustrates how academic research can be adapted for industrial testing, with the institute’s navigation models providing a basis for further development by the industry partner. ]]>