More Related Content
Similar to Autonomy and autonomous systems, the state of play
Similar to Autonomy and autonomous systems, the state of play (20)
Recently uploaded
Recently uploaded (20)
Autonomy and autonomous systems, the state of play
- 1. Introduction Current state Challenges Autonomy and autonomous systems, the state of play AI Nuclear Risk Workshop, SIPRI Dimitri Scheftelowitsch TU Dortmund May 22, 2018
- 2. Introduction Current state Challenges Autonomous systems Figure: Autonomous device warning, © Lifeboat Foundation What are autonomous systems?
- 3. Introduction Current state Challenges A definition Definition (An a empt) An autonomous system is a technical device that can
- 4. Introduction Current state Challenges A definition Definition (An a empt) An autonomous system is a technical device that can observe its environment (“Where am I?”)
- 5. Introduction Current state Challenges A definition Definition (An a empt) An autonomous system is a technical device that can observe its environment (“Where am I?”) plan own actions according to some goal (“How do I get what I want?”)
- 6. Introduction Current state Challenges A definition Definition (An a empt) An autonomous system is a technical device that can observe its environment (“Where am I?”) plan own actions according to some goal (“How do I get what I want?”) execute the planned schedule (“What do I do now?”) without (or with li le) intervention from a human operator.
- 7. Introduction Current state Challenges A definition Definition (An a empt) An autonomous system is a technical device that can observe its environment (“Where am I?”) plan own actions according to some goal (“How do I get what I want?”) execute the planned schedule (“What do I do now?”) without (or with li le) intervention from a human operator. Goal required!
- 8. Introduction Current state Challenges Notes on definition (a) DARPA 2007 Urban challenge participant (b) Spot welding robots, CC BY-SA 2.0 BMW Werk Leipzig Distinction autonomy vs. automation
- 9. Introduction Current state Challenges Notes on definition (a) DARPA 2007 Urban challenge participant (b) Spot welding robots, CC BY-SA 2.0 BMW Werk Leipzig Distinction autonomy vs. automation Automation execution of simple, hard-coded tasks
- 10. Introduction Current state Challenges Notes on definition (a) DARPA 2007 Urban challenge participant (b) Spot welding robots, CC BY-SA 2.0 BMW Werk Leipzig Distinction autonomy vs. automation Automation execution of simple, hard-coded tasks Autonomy algorithmic planning and scheduling
- 11. Introduction Current state Challenges Notes on definition (a) DARPA 2007 Urban challenge participant (b) Spot welding robots, CC BY-SA 2.0 BMW Werk Leipzig Distinction autonomy vs. automation Automation execution of simple, hard-coded tasks Autonomy algorithmic planning and scheduling Note distinction fuzzy
- 12. Introduction Current state Challenges Building blocks 0101… Any autonomous device needs sensors
- 13. Introduction Current state Challenges Building blocks 0101… ML Any autonomous device needs sensors internal domain representation
- 14. Introduction Current state Challenges Building blocks 0101… ML Any autonomous device needs sensors internal domain representation a model of own actions (“What if?..”)
- 15. Introduction Current state Challenges Navigation (a) DARPA 2007 Urban challenge participant (b) SailDrone, Photo © Richard Jenkins (c) Autonomous farming, Photo © Freya Fleckenstein (d) Cora, Photo © Ki y Hawk
- 16. Introduction Current state Challenges Medicine Most researched subject: Autonomous surgeries Figure: CyberKnife, CC BY-SA BLK Cyberknife Hospital Current stage CNC-like execution of surgical plan
- 17. Introduction Current state Challenges Medicine Most researched subject: Autonomous surgeries Figure: CyberKnife, CC BY-SA BLK Cyberknife Hospital Current stage CNC-like execution of surgical plan Cu ing edge Tumor recognition and radiotherapy
- 18. Introduction Current state Challenges Medicine Most researched subject: Autonomous surgeries Figure: CyberKnife, CC BY-SA BLK Cyberknife Hospital Current stage CNC-like execution of surgical plan Cu ing edge Tumor recognition and radiotherapy Future Autonomous cuts
- 19. Introduction Current state Challenges Non-robot applications Not always an autonomous device is a (mechanical) robot.
- 20. Introduction Current state Challenges Non-robot applications Not always an autonomous device is a (mechanical) robot. Example Autonomous trading
- 21. Introduction Current state Challenges Non-robot applications Not always an autonomous device is a (mechanical) robot. Example Autonomous trading Autonomous disaster warning
- 22. Introduction Current state Challenges Non-robot applications Not always an autonomous device is a (mechanical) robot. Example Autonomous trading Autonomous disaster warning Air traffic control
- 23. Introduction Current state Challenges Non-robot applications Not always an autonomous device is a (mechanical) robot. Example Autonomous trading Autonomous disaster warning Air traffic control Calling companies to update opening hours database
- 24. Introduction Current state Challenges Impossibilities Three (and a half) main problems Observation complexity
- 25. Introduction Current state Challenges Impossibilities Three (and a half) main problems Observation complexity Domain complexity
- 26. Introduction Current state Challenges Impossibilities Three (and a half) main problems Observation complexity Domain complexity Planning complexity
- 27. Introduction Current state Challenges Impossibilities Three (and a half) main problems Observation complexity Domain complexity Planning complexity Specification
- 28. Introduction Current state Challenges Observation complexity Not all observation tasks are equal Figure: German driver’s license exam, © fahrschule.de
- 29. Introduction Current state Challenges Complex domains: State space explosion Problem size is exponential in the number of variables 2 4 6 8 10 0 200 400 600 800 1,000 Variables States
- 30. Introduction Current state Challenges Complex domains II: Multiple parties and uncertainty Non-cooperative actors and uncertainty lead to difficulties Figure: 2010 Flash Crash Need to infer all goals of all actors and hidden information, otherwise: model unusable
- 31. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI…
- 32. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI… on May 22
- 33. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI… on May 22 departing in Bad Hersfeld, Germany
- 34. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI… on May 22 departing in Bad Hersfeld, Germany on May 22
- 35. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI… on May 22 departing in Bad Hersfeld, Germany on May 22 with a total budget of…
- 36. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI… on May 22 departing in Bad Hersfeld, Germany on May 22 with a total budget of… with at most 4 changes
- 37. Introduction Current state Challenges Planning Depending on the goal, planning can get complicated Example Get to SIPRI… on May 22 departing in Bad Hersfeld, Germany on May 22 with a total budget of… with at most 4 changes with 10kg luggage
- 38. Introduction Current state Challenges Problem formulation Not a computational problem… Problem Communication, especially with a computer, is non-trivial
- 39. Introduction Current state Challenges Problem formulation Not a computational problem… Problem Communication, especially with a computer, is non-trivial Formalizing a problem is often hard.
- 40. Introduction Current state Challenges Problem formulation Not a computational problem… Problem Communication, especially with a computer, is non-trivial Formalizing a problem is often hard. The wrong way World domination
- 41. Introduction Current state Challenges Problem formulation Not a computational problem… Problem Communication, especially with a computer, is non-trivial Formalizing a problem is often hard. The wrong way World domination The correct way Allocate shares to most profitable assets