HYDERABAD: What happens when robots begin to understand instructions like humans do, identify objects and carry out tasks without being given exact coordinates? What if they could also detect obstacles such as glass walls that their cameras fail to recognise?
These possibilities are taking shape at the International Institute of Information Technology Hyderabad (IIIT-H), whose researchers presented a range of innovations at the International Conference on Intelligent Robots and Systems (IROS) 2026 in Pittsburgh, the US.
The institute also scored big on the research front. It presented 10 research papers and one workshop paper, all of which were accepted — a 100% acceptance rate against the conference’s overall acceptance rate of less than 37%. Its researchers also took home the Best Paper Award at a workshop.
Consider AERMANI-VLM. Drones fitted with robotic arms can be useful for tasks that require objects to be located, picked up and moved. But getting them to understand what a person wants, rather than simply follow pre-fed coordinates, is a different challenge.
The framework enables these drones to interpret natural-language instructions and carry out such tasks. It also keeps the AI model’s decision-making separate from the drone’s physical movements, using a set of pre-approved actions to prevent unsafe or impractical commands.
Then there is the problem of glass. For a robot navigating a building or an unfamiliar space, a transparent wall can be a difficult obstacle to detect. GlassFormer addresses this by combining camera images with 60.5 GHz millimetre-wave radar. The system can identify glass surfaces even in poor lighting conditions. In tests, it recorded a detection score of 0.59, compared with 0.49 for a vision-only model.
IIIT-H develops framework to help robots map the unknown
The IIIT-H’s award-winning work explored another challenge: how can a robot build an accurate picture of an environment when it does not fully understand what it is seeing? The paper ‘Beyond Frontiers: Scene-Anomaly Guided Autonomous Exploration’, presented at the ‘Bridging Perspectives in Navigation: Lessons Learned and Future Directions’ workshop, proposed SCAGE, a framework that helps robots identify gaps in their understanding of an environment and explore those areas to build more complete and accurate 3D maps.
Authored by Akash Kumbar and Abhinav Raundhal under the supervision of Professor K Madhava Krishna, the paper won the Best Paper Award, which carried a cash prize of USD 500 sponsored by Field AI.
Another study turned to a challenge faced by drones operating in cluttered environments. Titled ‘AeroGrab: A Unified Framework for Aerial Grasping in Cluttered Environments’, it developed a system that allows aerial robots to examine objects from different angles and work out how to pick them up safely without colliding with surrounding objects.
The research extended to the coordination and control of aerial robots, too. Some studies explored improving their control through real-time learning, while others looked at calculating how many spare drones would be needed for inspection missions and using AI-based techniques to identify glass surfaces and estimate their depth.
A study titled ‘Robust Fleet Sizing for Multi-UAV Inspection Missions under Synchronised Replacement Demand’ also looked at the logistics of keeping drone fleets operational during inspection missions. They examined what happens when several drones need battery replacements at the same time. Across five test scenarios, the framework achieved 99.8% mission success by accounting for such simultaneous replacement requirements.
The work was not limited to aerial robots. At a workshop, IIIT-H researchers also presented a paper on combining radar and camera images to help autonomous vehicles estimate their movement and identify moving objects.
Describing the outcome as phenomenal, Professor Madhava Krishna said the institute had received an encouraging response to its research.
