When humans walk, they do not consciously compute muscle strength or body weight with each step. They adjust. That instinctive ability to adapt, rather than calculate, is now shaping a new line of thinking in robotics at the International Institute of Information Technology, Hyderabad (IIIT-H).
At the institute’s Robotics Research Centre, a team led by Prof Spandan Roy is developing drones that respond to changing conditions in real time. Instead of relying on rigid mathematical models, these systems are designed to adjust mid-flight to disturbances such as wind gusts, shifting payloads or sudden motion changes.
This marks a shift from conventional engineering approaches, where machines depend on precise models and predictable environments. In practice, however, real-world conditions are rarely stable. Drones operating outside controlled settings encounter uncertainties that are difficult to anticipate or encode in advance. The IIIT-H team’s work accepts this unpredictability and builds systems that can function despite it.
At the core of this approach is a simple premise: machines do not need complete information to perform effectively. Drawing from control theory, the researchers have designed systems that adjust behaviour on the fly, even when key parameters are unknown. Prof Roy compares this to human movement. “When we walk or lift objects, we do not calculate every variable, yet we adapt instantly,” he said.
One of the central challenges the team is addressing is what he refer to as ‘switching dynamics’. A common example is a delivery drone releasing a package. The moment the payload is dropped, the drone’s weight changes abruptly, affecting its motion. “Traditional systems often struggle to handle such transitions smoothly. The adaptive model developed at IIIT-H stabilises the drone almost immediately, allowing it to continue operating without disruption,” he highlights.
Handling instability
Another challenge arises when drones carry suspended loads. These loads tend to swing unpredictably, creating instability and safety risks. Instead of attempting to eliminate this motion altogether, the researchers focus on controlling it. By stabilising the system rather than the environment, the drone remains functional even under dynamic conditions.
The work also extends beyond navigation into aerial interaction. “We are exploring how drones can engage physically with objects while in flight. In one project, we developed a gripper inspired by slap bands that functions without motors, reducing mechanical complexity,” he said. In another, drones are being trained to catch objects mid-air using mechanisms that absorb impact without disturbing flight stability.
Such capabilities point to a clear shift in how drones may be used. Rather than serving only as carriers, they could take on roles that require handling, positioning or interacting with objects in real time. This has implications for sectors where precision and responsiveness are critical.
Practical applications
The research is already finding applications beyond the lab. In agriculture, the team has developed a drone-based pollination system that combines airflow with gentle physical contact to distribute pollen. “Field trials have recorded over 120 % effectiveness compared to conventional methods. In industrial environments, similar systems could automate object handling in warehouses. Meanwhile, reconfigurable drones that adapt to different payloads and missions are being explored for defence use, with some efforts leading to the creation of a startup,” he noted.
A notable feature of the lab is its collaborative structure. Despite a small pool of long-term researchers, students play a central role in hardware development, while responsibilities shift based on expertise. This model has resulted in consistent academic output, including nearly 30 journal publications over the past six years.
The transition from research to deployment is underway. With patents in progress and industry collaborations taking shape, the team is moving towards commercial applications. The agricultural system, initially developed as a sponsored project, is among the first expected to see wider use.
Yet the broader aim extends beyond individual use cases. The focus is on building systems that can operate reliably in uncertain and changing environments. Instead of waiting for perfect inputs, these machines are designed to respond, adjust and continue functioning, he explained.
In doing so, the work at IIIT-H points to a shift in robotics — from systems that follow instructions to those that adapt as they act.