added basckground chapter . First flush.

This commit is contained in:
Burkhart Wolff 2022-01-08 22:22:22 +01:00
parent 6889e08f33
commit b5939bc9db
2 changed files with 57 additions and 20 deletions

View File

@ -1298,6 +1298,37 @@ isbn="978-3-540-48509-4"
location = {Paris, France}
}
@article{arXiv170806374,
author = {{Shalev-Shwartz}, Shai and {Shammah}, Shaked and {Shashua}, Amnon},
title = "{On a Formal Model of Safe and Scalable Self-driving Cars}",
journal = {arXiv e-prints},
keywords = {Computer Science - Robotics, Computer Science - Artificial Intelligence, Statistics - Machine Learning},
year = 2017,
month = aug,
eid = {arXiv:1708.06374},
pages = {arXiv:1708.06374},
archivePrefix = {arXiv},
eprint = {1708.06374},
primaryClass = {cs.RO},
adsurl = {https://ui.adsabs.harvard.edu/abs/2017arXiv170806374S},
adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}
@techreport{wolff:hal03429597,
title = {{Modelling and Proving Safety in Autonomous Cars Scenarios in HOL-CSP}},
author = {Paolo and Taha, Safouan and Wolff, Burkhart},
url = {https://hal.inria.fr/hal-03429597},
type = {Research Report},
institution = {{University Paris-Saclay ; IRT SystemX, Palaiseau}},
year = {2021},
month = oct,
pdf = {https://hal.inria.fr/hal-03429597/file/document.pdf},
hal_id = {hal-03429597},
hal_version = {v1}
}
@Book{ nipkow.ea:isabelle:2002,
author = {Tobias Nipkow and Lawrence C. Paulson and Markus Wenzel},
title = {Isabelle/{HOL}---A Proof Assistant for Higher-Order Logic},

View File

@ -13,6 +13,7 @@ declare[[ Theorem_default_class = "theorem"]]
define_shortcut* csp \<rightleftharpoons> \<open>CSP\<close>
holcsp \<rightleftharpoons> \<open>HOL-CSP\<close>
hol \<rightleftharpoons> \<open>HOL\<close>
isabelle \<rightleftharpoons> \<open>Isabelle/HOL\<close>
(*>*)
@ -56,7 +57,7 @@ text*[introtext::introduction]\<open>
A widely accepted definition of Cyber-physical Systems (CPS) such as robots or autonomous cars
would comprise the following feature-list: they are systems involving
\<^item> concurrent interaction of digitally controlled actors and an environments,
\<^item> dense time (\<real>\<ge>0),
\<^item> dense time (\<open>\<real>\<ge>0\<close>),
\<^item> continuous environment with multi-dimensional observables
underlying the laws of (Newtonian) physics,
\<^item> the discretising view of continuous observables in
@ -74,9 +75,9 @@ Our understanding of 'analysis' of CPS involves:
In this paper, we present a \<^emph>\<open>framework\<close> for modeling and formally analysing CPS's and
a concrete \<^emph>\<open>instance\<close> in form of a case study from the autonomous car domain.
The latter is concerned with the formal verification of a particular collision-avoiding driving
strategy developed by @{cite "RSS Paper"}. Roughly speaking, our framework is based on a
conservative embedding called HOL-CSP2.0 @{cite "AFP entry"} into \<^isabelle> @{cite "XXX"}
which serves both as modeling and analysis environment.
strategy developed by @{cite "wolff:hal03429597"}. Roughly speaking, our framework is based on a
conservative embedding called HOL-CSP2.0 @{cite "HOL-CSP-AFP"} into \<^isabelle>
@{cite "nipkow.ea:isabelle:2002"} which serves both as modeling and analysis environment.
Describing our foundational stack in more detail, we'd like to mention:
\<^item> Communicating Sequential Processes (\<^csp>), which is a language
@ -97,17 +98,20 @@ Describing our foundational stack in more detail, we'd like to mention:
\<^emph>\<open>conservative extensions\<close>, comprise a notable amount of number theory, real analysis, and
program semantics theories.
\<close>
text\<open>
The expressive power of \<^holcsp> stems from the fact that its \<^emph>\<open>channels\<close> may comprise arbitrary
\<^hol>-types, and thus infinite data for which powerful interactive and automated proof techniques
are available. In particular, CPS and \<^holcsp> fit particularly well together since:
% better a presentation as table ?
\<^item> concurrent interaction of actors and environments:
this is what \<^csp> is built for;
\<^item> dense time (\<open>\<real>\<ge>0\<close>): the libraries provide real numbers as well as rich theories to reason
about them;
\<^item> continuous physical environment with multi-dimensional observables
underlying the laws of (Newtonian) physics: the Isabelle/HOL-Analysis library provides
definitions and proof techniques to reason over differential equations;
\<^item> concurrent interaction of actors and environments: this is what \<^csp> is built for;
\<^item> dense time (\<open>\<real> \<ge> 0\<close>):
the libraries provide real numbers as well as rich
theories to reason about them;
\<^item> continuous physical environment with multi-dimensional observables underlying the laws of
(Newtonian) physics: the Isabelle/HOL-Analysis library provides definitions and proof
techniques to reason over differential equations;
\<^item> discretized/sampled view of continuous variables:
\<^hol> libraries provide theories for vector-spaces for positions, forces, accellerations, etc,
as well as (low-level, machine-oriented) bitvector calculations
@ -115,20 +119,22 @@ are available. In particular, CPS and \<^holcsp> fit particularly well together
can be adapted, as we will see.
\<close>
text\<open>This paper proceeds as follows:
after an introduction into \<^holcsp> we present the general framework of
interacting autonomous cars, which may be modeled in a object-oriented manner as
in the domain-specific ontology language Mozart @{cite "XXX"}. This involves the conception
of an orchestrator (the 'maxwell-daemon') that represents the assumption that all actors
posses "perfect" captors and 'know' at some points in time the absolute physical state
of all other actors, and that react on this knowledge by a particular non-deterministic
\<^emph>\<open>driving strategy\<close>.
text\<open>This paper proceeds as follows: after an introduction into \<^holcsp> we present the general
framework of interacting autonomous cars, which may be modeled in a object-oriented manner as
in the domain-specific ontology languages such as
MOSAR \<^footnote>\<open>\<^url>\<open>https://www.mosar.io\<close>\<close> or
Foretellix's M-SDL \<^footnote>\<open>\<^url>\<open>https://www.foretellix.com/open-language/\<close>\<close>.
This involves the conception of an orchestrator (the 'maxwell-daemon') that represents the
assumption that all actors posses "perfect" captors and 'know' at some points in time the absolute
physical state of all other actors, and that react on this knowledge by a particular
non-deterministic \<^emph>\<open>driving strategy\<close>.
Subsequently, inside this framework, we present a concrete topology and scenario with
two or more cars on a lane, that follow the RSS driving strategy. The latter has been
proposed by S. Shaliv-Shwartz, S. Shammah, A. Shashua:
"On a Formal Model of Safe and Scalable Self-driving Cars"
@{cite \<open>XXX\<close>} and which comes up with a concrete formal definition
\cite{arXiv170806374} and which comes up with a concrete formal definition
compatible with the law ("Duty of Care"). Moreover, the authors provide
a paper-and-pencil proof trying to establish a global safety property over a class of scenarios,
which basically excludes the existence of collisions.