MURI Telecon 2005-07-13
From Murray WikiJump to navigationJump to search
- Quick review of RFP
- Round robin: comments on RFP, interests, etc
- Wrap up and next steps
- V&V is really messy and hard the way that it is currently done
- Experience working with Boeing/OCP isn't encouraging
- How do you make code for a distributed system in a straightforward way that is automatically checked, etc?
- How do you know the way the controller is switched on/off is worked into the system?
- Systems Eric can get his head around are much simpler; agents running a controller, switching based on messages, etc. Can put into formal setting and reason about it; not clear how to reason about the dynamics
- Interested in figuring out how to build these sorts of systems using structured principles
- Focus on basic science - create theoretical infrastructure that addresses all of the research concentration areas
- Start with Prajna thesis - V&V of hybrid systems
- Build this into the SoS framework (Pablo can elaborate)
- Restrict the expressiveness of the languages (protocols, not systems). Restrict the kind of hacks people can do.
- Have examples of distributed systems using protocol stacks - use this sort of an architecture
- Think about the mathematics
- Doing things bottom up (starting with code that someone has developed) is impossible
- Start top down: what's the language for specification (temporal logic, some variant?)
- What is the theory behind the specification
- SoS + temporal logic
- Add Gerard Holzman (JPL), Rajiv Joshi (JPL)?
- Specification, theorem proving, etc
- Try to stay out of things like CORBA and other large systems that people have built up
- Use of probability: can we prove things things in a probabalistic manner?
- Particularly interested in how we integrate discrete and continuous parts
- One possibility is establishing a middle layer where we can assert things about the time evolution (certain things will transition in a period of time)
- Barrier work of Steven could be a start; is there a way to integrate things in a tighter way?
- Incorporate adversarial effort?
- Erik: moving back and forth between writing continuous part using discrete language versus the other way
- Common way of thinking about: polynomial representations for discrete problems. But these are extremely structured proofs.
- Possible key themes:
- Problem features: hybrid, distributed, stochastic, adversarial
- Fundamental framework for reasoning about distributed computation for real-time, feedback systems
- Top-down approach: think about a language for specifications that we can reason about
- Think about whether we can incorporate probability in a formal way
- Should we add others to the group?
- Put together set of 3-5 papers describing the basic ideas that we want to build on
- CCL paper
- V&V from Mani
- SoS Barrier Certificates