A robot developed through a collaboration between the International Institute of Information Technology Hyderabad (IIIT-H) and Ati Robotics has navigated to a target on a factory floor using camera images and artificial intelligence-based 3D reconstruction, without relying on conventional LiDAR maps.
The demonstration tested research developed by IIIT-H’s Robotics Research Center as part of a partnership with Ati Robotics that began in 2023. The research examines whether robots can use visual cues, recognisable landmarks and spatial relationships to locate themselves and plan routes, in a manner similar to how people use familiar features to navigate.
LiDAR-based systems typically map their surroundings and help robots estimate their position through simultaneous localisation and mapping, or SLAM. The IIIT-H approach instead represents an environment through a smaller set of landmarks. In a warehouse, these may include areas near a door or opposite an elevator, rather than detailed measurements of every part of the facility.
Prof. Madhava Krishna of the Robotics Research Center said the team investigated whether images and artificial intelligence could allow robots to navigate without building dense, detailed maps. The researchers also tested a method in which an operator identifies a destination by pointing to a pixel in an image instead of entering precise coordinates.
Under this method, the system compares images to identify corresponding points in the real world, estimates the spatial relationships between them and assesses possible routes. The demonstration was conducted on a factory floor, while the research has been developed for practical use in warehouse robots.
Prof. Sourav Garg of the centre said the work had resulted in papers presented at international forums including the IEEE International Conference on Robotics and Automation, NeurIPS and the Conference on Robot Learning. IIIT-H has transferred code to Ati Robotics for testing and deployment in warehouse robots.
Ati Robotics has adapted parts of the research into OccupancyReact, a LiDAR-based variant that is closer to its existing systems. The company said its current focus was on improving the final stages of robot movement, including docking, approaching pallets and manoeuvring through narrow spaces.
These tasks are operationally significant in warehouses because small reductions in movement time can accumulate across a working shift. The collaboration therefore links university research in visual navigation with the practical requirements of industrial robot deployment. The next stage is testing and adapting the transferred code within warehouse systems, while Ati Robotics continues work on docking, pallet approach and movement through confined spaces.
