PRE2018 3 Group17/Old
This section will be rewritten as mentioned in the new planning
It is important to first find out what has already been done on the subject matter. This is done by looking at the state of the art research done for these rescue drones. These can then be used to either help develop the proposed solution, or to use research as an additional component that works in tangent to the solution.
Existing components of the design
The fact that drones have grown massively in popularity over the past decade is clear, but scientists have shown increased interest in drone technology as well [11]. Drones and other UAVs (Unmanned Aerial Vehicles) are now used in a wide variety of applications including the making of movies, surveillance and inspection of industrial structures that take up large amounts of space (oil pipelines, train tracks), mapping geological structures and providing food and medical relief to hard to reach areas [10]. The Delft University of technology recently developed an ambulance drone which carries several useful pieces of equipment over to the specified location, so people can get a head start on helping the person in need [16].
Understanding of how to effectively use drones in different kinds of applications is increasing rapidly, and the subject of using drones in disaster areas is no exception. Research shows that as drones became more affordable, and their technology more advanced, they are becoming increasingly suitable for implementation into disaster areas. Even if rescue personnel are not adept at using the technology, it still manages to increase their efficiency [4]. For instance, the usage of drones as cellular beacons in case cell towers no longer function is being investigated [14], and drones are being used for support and observation, including providing information about the development of forest fires, and information about collapsed buildings [5].
One part of the problem of using drones in disaster areas, namely the autonomous analysis of the images captured by the drones, has been studied extensively. Systems have been developed for detecting potential obstacles for the drones [13], recognizing to what degree a building has collapsed [4], or how to recognize different types of forest fires [15]. Similar systems exist using observation from space by satellite, but these methods often lack the resolution required to get all the information required [17]. The recognition of humans from these camera images has also been studied. This is done either by teaching a system to recognize humans from a set of test images [24], or by comparing heat signatures [26]. Since these technologies are already quite well researched, we will focus on other parts of the project, namely optimizing the search strategy of the drones.
Pathfinding through rough terrain
Path planning for robots can be done in multiple ways, but finding the right choice for rescue operations can prove cumbersome. Research has been done to improve the path planning in regards to time planning and determining if the path taken can be completed. One such research is using genetic algorithms to determining a path as shown in source [fuzzy evolutionary algorithms]. This does however have the drawback as it assumes to know what terrain is difficult to traverse and what isn't. However this method is able to deal with unexpected situations and plans a new path that is close to optimal to reach its goal.
Another method is to evaluate the chance the robot will tilt when moving through the disaster area [attitude maneuver]. This is done by determining the height of the area using sensors, and constructing a height gradient. The robot can then decide on a path through this gradient after nodes have been set, it takes into account the length of the path and the chance of tilting over. This method is ideal for small case areas, but would need some considerable computation power to reliably do this continuously.
A variation on this idea is to change the configuration of actuators depending on the terrain [Reconfigurable robots]. This combines the path planning of the previous idea with additional functionality to further decrease the chance of tipping over. This can therefore be added as an extra to existing robots, given that it knows what the path will be like when moving towards it.
Using deep reinforcement learning is also an option for terrain navigation [Reinforcement learning]. This method uses an elevation map as well, and can learn what route it should take to reach the goal. This can then be applied to a robot and it should select a a succesful route, it could even learn if it makes a mistake. This aspect of self improvement is unique to deep learning.
The final option that was researched is the option of using a guidance system that will guide a robot through dangerous areas [Guidance]. This guidance system can use a multitude of lightweight sensors that can be placed all around the area. These will then connect with the main network and determine what areas are hazardous. This system is not ideal to move around obstacles. It is however useful in finding survivors as this is another functionality of this design.
Useful resources for the design
- 1 Talks about how drones can autonomously find survivors by scanning the environment. They offer a high potential for fast and efficient response during a rescue mission. What should the drone do to help the survivors. Needs to observe its environment to avoid a collision.
- 7 Talks about the feasibility of a multi-tier drone architecture over single tier drones in terms of efficiency. This increases efficiency and reduces path loss.
- 12 Mainly talks about how paths are found for drones to follow. and how trajectory planning works, uses decision making and direction of target given a path to deciding what to do.
- 14. Talks about the usage of drones as cellular network beacons in cities after some calamity. Presents a stochastic model to predict how many drones are necessary for a given situation. We could do something similar for our number.
- 18. Proposes HAC-ER, a system for cooperation between information-gathering agents and humans in disaster regions. Shows great promise, problems mainly arise due to airports not easily allowing UAVs in their active airspace.
- 22. Article discusses the benefits and drawbacks of two different communication methods for drone swarms, also explains them. Very useful and relevant for later stage of our project.
- 23. Article acts as an example of how to effectively set up a communication network for a 'swarm' of agents, how to get them to perform tasks. Super useful, but hard to follow.
- 25. Design for a fully autonomous/wireless drone charging station. Useful if we want to include a charging station in our strategy.
Complementary sources
- 1 Talks about how drones can autonomously find survivors by scanning the environment. They offer a high potential for fast and efficient response during a rescue mission. What should the drone do to help the survivors. Needs to observe its environment to avoid a collision.
- 7 Talks about the feasibility of a multi-tier drone architecture over single tier drones in terms of efficiency. This increases efficiency and reduces path loss.
- 14. Talks about the usage of drones as cellular network beacons in cities after some calamity. Presents a stochastic model to predict how many drones are necessary for a given situation. We could do something similar for our number.
- 2 Discusses failure of a drone system, espionage due to hacking, and autonomous finding of survivors. (Weinig text om er meer over te zeggen)
- 8 Discusses the useful sections of implementing drones in rescue scenarios, as well as how to manage certain aspects of it. It gives a summary of various communication aspects and issues related to their deployment.
- 9 Discussion on why opportunistic networks aren't more common in todays world
- 11 Optimization approaches for different civil applications of drones and characteristics of those types of drones, drones are extremely versatile and new uses are always found for them.
- 20. Article discusses general ethicalness of CCTV surveillance. Concludes that partially automated data analysis from these systems is more ethically preferable to manual analysis. Hard to relate to our case specifically, but could spark the discussion of ethicalness of our system.
- 21. Article discusses ethical feasibility of facial recognition systems. Seems unrelated to our project entirely.
Inaccessible sources
- 3 Presents a vision where the drones provide wireless communication between survivors and cellular infrastructure. (Geen toegang tot volledige artikel)
- 19. This article is not accessible using a TU licence. Abstract talks about integrating calculations for path planning between different scales of a system (i.e. destination of each agent vs. not crashing into each other etc.).