Danesh Tarapore

Autonomous robots operating in the real world will inevitably encounter situations that were not anticipated when they were designed. Hardware can fail, terrain can defeat a locomotion strategy, robots can become physically trapped, and environmental conditions can invalidate previously successful behaviours.

This theme investigates how robots can recognise, diagnose and recover from such situations autonomously, rather than relying on a human operator to intervene. The work ranges from rapid behavioural adaptation following damage in individual robots, to fault recovery in robot swarms, and more recently to robots that can detect and recover from entrapment in complex natural environments. A recurring idea is that resilience should come from the robot’s ability to exploit its own experience and behavioural capabilities to find alternative ways of achieving its objectives.

The long-term goal is autonomy that does not simply work when the world behaves as expected, but can continue to function when it does not.

Forests and other natural environments are difficult to observe at the spatial and temporal scales needed to understand how they are changing. Ground measurements provide rich information but are labour-intensive and geographically sparse, while aerial and satellite observations cannot capture everything occurring beneath the canopy.

This theme explores the use of autonomous mobile robots as environmental sensing platforms, particularly for forest monitoring. Robots can move through an environment, position sensors where measurements are needed, revisit locations over time, and collect observations from perspectives that complement established remote-sensing techniques. The research addresses both the sensing problem and the autonomy needed to make such measurements practical: locomotion through difficult terrain, localisation beneath vegetation, low-cost and resource-efficient perception, and coordination between multiple robots.

The longer-term ambition is to develop robotic systems that become part of the environmental scientist’s toolkit, providing persistent, scalable and spatially detailed observations of complex ecosystems.

A robot swarm gains its power from having many relatively simple robots cooperate without requiring a single robot or central controller to understand the entire system. But real robotic collectives cannot assume unlimited communication, computation, energy or sensing. These constraints fundamentally affect what a group can know and what decisions it can make.

This theme studies how useful collective behaviour can emerge despite those limitations. The research includes decentralised decision-making, allocation of robots to competing alternatives, adaptation when members of the collective fail, and methods for reducing the computational and communication resources required for collective intelligence. Rather than treating resource constraints simply as an engineering inconvenience, the work asks how they change the design of the algorithms themselves.

The aim is to understand how large groups of robots can remain scalable, distributed and effective under the constraints encountered outside the laboratory, from environmental monitoring on land to robotic operations at sea.