The Robotics Research Center (RRC) at the International Institute of Information Technology Hyderabad (IIIT-H) demonstrated a vision-based navigation system developed through a three-year collaboration with Ati Robotics. The technology enables autonomous mobile robots (AMRs) to navigate factory environments using cameras, visual data and artificial intelligence (AI), offering an alternative to conventional LiDAR-based mapping approaches.
During a recent demonstration at Ati Robotics’ factory, an AMR successfully navigated to a designated object using images captured during an earlier walkthrough. The system employed a 3D foundational model to derive geometric information from pairs of images. Unlike conventional navigation systems, this demonstration did not depend on LiDAR occupancy grids, scan-matched positioning or metric maps.
Exploring Landmark-Based Robot Navigation
Traditionally, robots use LiDAR sensors to map their surroundings and estimate their positions through simultaneous localisation and mapping (SLAM). However, the IIIT-H research team explored an alternative approach that relies on visual information and recognisable landmarks to guide robot movement.
The system uses sparse localisation to represent an environment through a limited number of reference points. Outdoors, these landmarks could include buildings, traffic signals and trees. In warehouses and factories, robots could recognise doors, lifts and designated storage areas to estimate their locations and navigate between destinations.
Explaining the approach, Prof. Madhava Krishna highlighted how humans use visual cues, memory and movement to navigate their surroundings. Rather than relying on precise three-dimensional coordinates, people continuously adjust their routes, avoid obstacles and respond to changing conditions. The research aims to bring similar adaptability to autonomous robot navigation.
Image-Based Navigation Guides Robot Movement
The team also investigated image-based navigation, which directs robots towards specific points in images rather than relying exclusively on precise spatial coordinates. In this approach, an AI model compares two images of the same environment to identify corresponding points and estimate their three-dimensional positions. It then generates a cost image that helps the robot evaluate possible movement directions and select an appropriate path.
Consequently, the robot can navigate towards specific destinations, such as pallets and docking stations, without depending on conventional three-dimensional coordinate maps. This approach could simplify navigation in industrial environments where visual landmarks provide sufficient information for movement and positioning.
Research Advances Towards Industrial Deployment
The research team has developed a navigation framework that integrates mapping, localisation, planning and control through the relative connectivity of image segments. “We built a navigation stack that defines mapping, localisation, planning and control simply through relative connectivity of image segments,” said Prof. Sourav Garg of the Robotics Research Center. The research has been presented at leading robotics and AI research forums, including the IEEE International Conference on Robotics and Automation, the Conference on Robot Learning and the Annual Conference on Neural Information Processing Systems.
Ati Robotics Adapts Research for Warehouse Applications
IIIT-H has transferred the research code to Ati Robotics for further testing and potential deployment in warehouse robotics. The company has also adapted elements of the research into OccupancyReact, a LiDAR-based variant designed to complement its existing deployed solutions. Prof. Garg explained that the navigation framework’s modular design enables researchers to improve their RGB-only navigation approach while allowing Ati Robotics to integrate selected components into its own technology stack. This flexibility could help accelerate the development of autonomous mobile robots for industrial applications. As reported by manufacturingtodayindia.com, by combining visual intelligence with adaptable navigation architectures, the collaboration aims to advance robotic mobility and support more flexible automation across factories and warehouses.
