Orchid Project

Effective disaster response requires rescue services to make critical decisions in the face of an uncertain and rapidly changing situation. We aim to develop systems that allow first responders and software agents to work effectively together in such situations to collect the best possible information from the environment (though diverse sources such as CCTV feeds, UAVs and crowd generated content), in order to most effectively manage and coordination the various rescue resources available.

Key technologies to achieve these aims include

1. Decentralised coordination algorithms that can effectively allocate resources in the absence of centralised control.
2. Methodologies to flexibly handle autonomy so that the decisions that are autonomously made by software agents can be continuously changed as needs arise.
3. The ability to track the provenance of information and decisions such that previous decisions can be updated as new information comes to light.

Learn more about Disaster Response Technologies

RCUK_Robotics_and_Autonomous_Systems_Timeline

 City-Wide Disaster Response in 2020 vignette

A multi-site sporting event is taking place across the country. As part of this, a large number of people travel to London for a major tennis tournament at the O2 arena in Greenwich and a beach volleyball event at the ExCeL exhibition and conference centre in Docklands. The build up to the event has coincided with stormy weather over the North Sea which has resulted in week-long downpours that have caused a number of flash floods (mainly due to blocked drainage and sewers) around the city. Furthermore, this stormy weather has generated a storm surge along the east coast and since this will coincide with the normal spring tide, the Thames Barrier has been closed as a precaution. However, these unusual circumstances have not deterred fans and as a result transport systems are congested: there are traffic jams at all major road arteries, all trains converging to the main sites are packed to their limits and the number of boats navigating the Thames in the centre of London is at record-breaking levels.

To guarantee security, police and security services are on alert. However, despite all the precautions taken, a number of explosive devices go off to the east of the city. The first of these devices targets the Thames Barrier. It damages the northernmost pier and punches a hole in the floodgate. The second and third devices go off near the entrance to the O2 arena and in the car park of the ExCeL centre. The force of water due to the storm surge and the high tide rapidly destroys the already weakened floodgate and the Thames overflows its bank in Greenwich, causing parts of the O2 arena to collapse. A major fire starts to spread around the parking lot outside the ExCeL centre, creating large plumes of dark smoke that can be seen for miles around. Crowds throughout the city start evacuating the area, afraid of further explosions and trying to avoid the rising water from the Thames. Within minutes of the event, live video is appearing on social networking sites around the world from people on the ground uploading data from their mobile phones.  Conventional media is much slower to respond. While there are journalists and TV Crews at the events, they have no specific information to add. As the emergency progresses, the mobile networks are rapidly overwhelmed by the peak in data traffic. As the emergency responders rush toward the scene of the emergency, their response is hampered by the breakdown of existing communication networks and the paucity of information about what is happening on the ground.

Flexible Autonomy

Such disasters require the use of Unmanned Ground Vehicles (UGVs) and Unmanned Aerial Vehicles (UAVs) in the surveillance, search and rescue operations, while software agents will assist with the job of assigning tasks to emergency responders based on the status and priority of casualties. To ensure these complex tasks are efficiently (maximising lives saved) and effectively completed human operators will have to work closely with the teams of autonomous vehicle and task allocation agents, continually varying the degree of autonomy given to them as the situation evolves. Given this, computationally efficient techniques are needed to determine when and to what extent human intervention is required to improve the performance of autonomous vehicles and agents. Mechanisms must be designed for the agents to detect and understand the actions and motives of their human counterparts in the response effort to ensure that they can engage in fruitful collaborations. It is also important that human factors be taken into account in presenting the controls for teams of vehicles and agents to ensure the operator understands the effects of their actions.

To address the challenges related to the development of flexible autonomy, some initial work in collaboration with the Australian Centre for Field Robotics has been focused on the development and the deployment of decentralised coordination techniques on real Unmanned Aerial Vehicles (UAVs).

Agile Teaming

As the disaster unfolds new tasks will appear and others will disappear, while the environmental conditions may change significantly from one moment to the next. HACs will have to be formed, through peer-to-peer negotiations between actors, at particular points in the disaster space, perform and complete some tasks and move onto new tasks that are deemed achievable and have a priority in the overall disaster response strategy. These teams may have to work together and maintain a high level of situational awareness from surveillance teams composed of heterogeneous agents (e.g. CCTV, UAVs and crowd generated content). The formation, disbanding and reformation of such teams will have to be fast and robust to uncertainty (e.g. weather conditions, flooded areas, traffic conditions) which could, as new information is obtained, invalidate previous actions chosen by the teams.

Incentive Engineering

By having a high degree of situational awareness, both from the start and throughout the disaster, the response effort can be efficiently organised to mitigate its impact. It is crucial that as much relevant information, in the most important areas, is gathered from the scene. This may be achieved using sensors located in the environment, or more importantly, using participatory sensing. This means that incentives and indirect control mechanisms must be provided to people to continuously generate updates about their state and surroundings to teams of rescue agents, providing situational awareness services (in the cloud) to emergency response agencies about their location and surrounding conditions. These incentives should encourage those people with useful information and deter those with data that provides little additional information. Equally, since the movement of citizens during a disaster is heavily dependent on the information provided to them about the dangers and safe spots in the disaster space, the throttling of information to selected groups of citizens may help to prevent chaotic evacuation and help reduce traffic jams during rescue missions.

Accountable Information Infrastructure

As several, possibly correlated, events occur before and during a disaster, a significant amount of redundant and possibly diverging reports will be propagated throughout the system. This information will be provided by a large number of mobile sources (humans, sensors and agents) in short-lived bursts and from fixed sensors (e.g. CCTV) with different degrees of accuracy and reliability. During the response effort a large number of decisions made by humans, autonomous vehicles and agents (e.g. to rescue citizens as opposed to extinguishing fires) with varying consequences (casualties die as a consequence of the fire spreading) may be taken and these need to be tracked in order to ensure incident commanders understand the effect of their choices and therefore help them adjust their strategy. To ensure the system is accountable and can be trusted by responders and incident commanders, it is important to devise computationally efficient techniques to track and visualise the provenance of large numbers of decisions and vast amounts of data, that can cope with incomplete and delayed data, coming from multiple, correlated, unverifiable and unreliable sources.

agents | citizen-science | applications | crowdsourcing | disaster response | smart grid | accountable information architecture |agent-based computing | agile teaming | disaster recovery |flexible autonomy | decentralised control | energy management |HACs | human-agent interaction |human computer interaction | human agent collectives | incentive engineering | machine learning |mechanism design | ORCHID |