1 / 22

Project Presentation: ‘Modelling Navigtion and Landmarking in Ants ’

Project Presentation: ‘Modelling Navigtion and Landmarking in Ants ’. Project by Daniel Keer Supervised by Dr Steve Russ Abstract:

aitana
Download Presentation

Project Presentation: ‘Modelling Navigtion and Landmarking in Ants ’

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Project Presentation: ‘ModellingNavigtion and Landmarking in Ants’ Project by Daniel Keer Supervised by Dr Steve Russ Abstract: An attempt to model some of the ways ants navigate, particularly focusing on their visual navigation methods and use of landmarks. The project has been undertaken from an Empirical Modelling perspective, using the modelling tool tkeden.

  2. Project Aims • Broadly: To model the navigational aspects of ant behaviour, from an Empirical Modelling perspective. • I wanted there to be a strong correlation with real scientific experiments and understanding. • I wanted to make the model as visual as possible, and for the workings of the AI to be transparent (in keeping with the Empirical Modelling perspective).

  3. Ants • I chose ants as a subject for the project because of the varied and quite amazing way that they navigate. • The ant problem: to forage for food and return to the nest as quickly and effectively as possible. • Ants have a range of sensory abilities and orientation methods that they use to do this (varying from species to species of course) • I will describe some of these briefly so that you can understand the model

  4. PATH INTGRATION Ants can keep track of the distance and direction of their return path to the nest when out foraging. This ability is called ‘Path Integration’. However their calculations are based on rough estimations (not integration), and are not totally reliable. My model implements a calculation method that ants are thought to use, and makes clear the discrepancy with the exact values.

  5. Path Integration When Foraging Black – outward path. red – return path You can see from this example foraging walk that an ant’s return path is accurate and (fairly) straight-line. This is despite a long and rambling outward search path.

  6. VISION Most species of ants have a keen sense of vision. In the model, I have divided the way they use sight into two distinct ways: • To recognise objects of interest nearby • – food, enemies, nestmates, etc. • 2. To recognise ‘landmarks’ • – taller objects such as trees, rocks, etc. Adjacency view Obstacle view

  7. VISION – LANDMARKING & SNAPSHOTS Many ant species are capable of remembering ‘snapshots’, which help them to remember the route to food. A snapshot is thought to be a representation of surrounding landmarks (focusing on the colour and elevation of tall objects). The procedure I will describe next is a simplified view of how ants combine their Path Integration and Snapshot abilities to navigate to food sources. Snapshot ‘memory’ in the model

  8. COMBINED ANT NAVIGATION SCENARIO When the match is good enough, it will move onto the next shot in memory, and will do the same again. After a set number of steps, the ant will stop, turn 180o and take a ‘snapshot’. It will fix any nearby objects in its memory. The ant will then head in the direction that its Path Integration record tells it is the way back to the nest. It will then try to find its way back to the food source. It will compare the ‘snap’ in memory to surroundings and move that way. This will be repeated several times on the way back to the nest. At the nest, it will deposit its food. …and may eventually be lucky enough to find some It will then begin a (fairly random) search for food… The ant starts off at the nest In this way, it can approximately relocate the food source

  9. Specific Project Aim • The scenario just described is an interesting problem in AI. • Modelling this type of complex behaviour became a main aim of the project. • Two major components are necessary for this: • A suitable environment and interface • (landmarks, ant, food, senses, etc) • The AI for navigation

  10. Basis for the Environment (K.C. Tan’s Model) • This Original Model Provided: • Toolbar • A Clock • Basic map • Ability to place blocks • Ranger • Ant and ‘rays’ • Food and scent (which I removed) I was lucky enough not to have to start from scratch with the environment. The environment that I developed was built on top of a previous project by K.C. Tan. This saved me a lot of time, and meant that I could implement some interesting features.

  11. Basis for the Environment (K.C. Tan’s Model) The changes that I have made to this original will be evident during the demonstration.

  12. AI MODES Return (Take Snapshots) Locate Nest • For the AI, I implemented a set of AI ‘modes’ that correlate with what the ant does under different situations – search, return to nest, etc. • The ant will switch between these modes under certain circumstances (e.g. if it finds food). • The mode that the ant is currently in is clearly displayed in the interface. Close (By P.I.) Return (Simple) Pick UpFood Pick UpFood Nest Found Travel Limit Exceeded At Nest Food Found Initiate Search Search Trail to Follow Lost Follow Route

  13. AI MODES Return (Take Snapshots) Locate Nest Close (By P.I.) Return (Simple) Pick UpFood Pick UpFood Nest Found Travel Limit Exceeded At Nest Food Found Initiate Search Search Trail to Follow Lost Follow Route

  14. AI MODES Return (Take Snapshots) Locate Nest Close (By P.I.) Return (Simple) Pick UpFood Pick UpFood Nest Found Travel Limit Exceeded At Nest Food Found Initiate Search Search Trail to Follow Lost Follow Route

  15. AI MODES Return (Take Snapshots) Locate Nest Close (By P.I.) Return (Simple) Pick UpFood Pick UpFood Nest Found Travel Limit Exceeded At Nest Food Found Initiate Search Search Trail to Follow Lost Follow Route

  16. AI MODES Return (Take Snapshots) Locate Nest Close (By P.I.) Return (Simple) Pick UpFood Pick UpFood Nest Found Travel Limit Exceeded At Nest Food Found Initiate Search Search Trail to Follow Lost Follow Route

  17. AI MODES Return (Take Snapshots) Locate Nest Close (By P.I.) Return (Simple) Pick UpFood Pick UpFood Nest Found Travel Limit Exceeded At Nest Food Found Initiate Search Search Trail to Follow Lost Follow Route

  18. KEY ACHIEVEMENTS IN THE AI • IN THE MODEL: • The ant is able to record snapshots into memory. • It can then follow the route back, comparing the snapshot with what it can currently see to decide how to turn. • It is usually quite successful at this (provided certain conditions are met regarding the surrounding objects). • This is the key achievement of my project.

  19. Design Process • For the environment/interface, I had a clear plan of what was necessary, and developed KC’s model accordingly. • When it came to the AI, I had some ideas about how I might implement the snapshot matching, but was unsure how successful these ideas would actually be in practice. • I experimented with these ideas in the model to see what behaviours would result. • I also generated new ideas by moving the ant myself and seeing what she could perceive of the environment. • I built up the complete working AI through this interaction and experimentation with the model.

  20. State of Completion • The environment and interface is pretty much complete. • The AI is complete enough to showcase the navigation method that I have modelled. • There are situations where the AI will not work perfectly (e.g. if there are objects of the same colour), and it is true that a lot more work could be done to fine-tune the AI perfectly (you could go on forever). • However, I think that the interesting part of this project has been showing the complex and lifelike behaviours that can result from a simple set of basic rules. • A ‘bulletproof’ AI is less important.

  21. Empirical Modelling Approach to AI • I think that the Empirical Modelling ‘experimental’ approach has significant merit for AI development. • The snapshot recording and following parts of the AI were the most difficult to develop. • Experimenting with the model from the ‘point of view’ of the ant was vital for generating ideas about how the ant could compare snapshots with current surroundings. • I think it would be very difficult to develop such an AI without ‘playing’ with the model to generate such insights • The Empirical Modelling approach seems a very sensible one to take for these types of problems.

  22. - END OF PRESENTATION (PART 1) - I will now demonstrate the model working before I attempt to explain anything more about it. (Feel free to ask any questions as I go along)

More Related