[ecoop-info] PhD position: Self-Adaptation and Resilience for Artificial Intelligence-Based Robotic Systems
ansgar.radermacher at cea.fr
Mon Apr 29 17:30:46 CEST 2019
The LSEA (Embedded and Autonomous Systems Design Laboratory) at CEA LIST
<http://www-list.cea.fr/en/talents-list/did-you-know> offers a PhD
Robotic systems have to integrate more and more functionality including
autonomous decisions how to adapt to changing environment conditions or
failures. These systems have to respect classical requirements of
embedded systems (resource, timeliness), but resilience to failures and
safety requirements become very important. Stopping the system is not an
option for instance for an autonomous vehicle or a drone, systems have
to be fail-operational. Another aspect is the use of AI components
(machine learning) in control algorithms and for taking autonomous
decisions - these systems are attractive, as they are able to abstract
and generalize by inductive inference. Whole they can offer excellent
performance in nominal conditions, the validation of such
decision-making system is complex. Even simple algorithms can create
emergent behaviour when systems interact and when feedback loops are
introduced. This raises the question how to validate the system
behaviour and its ability to anticipate, resist, reconfigure itself
after a fault or interference.
Autonomous systems must adapt to changing environmental conditions.
Although the new configurations can be checked offline, this approach is
not flexible enough, it only applies to the configurations planned at
development time and does not scale because the number of potential
configurations increases exponentially with the number of components (as
examined in the European project SafeAdapt (www.safeadapt.eu). Thus, the
system has to check at runtime whether new configurations meet the
functional and non-functional constraints. The use of a model @ runtime
is a way of attacking this problem.
The objective of this thesis is to embed validation and reconfiguration
mechanisms into the running system while specifically taking AI
applications into account that learn and evolve. A promising approach is
resilient engineering which uses models as run-time central elements
providing the opportunity to learn from past events, plan, and recover
from a dangerous decision.
As the subject is very broad, we propose to work from the bottom up with
case studies that seem representative in the field of collaborative
robotics and autonomous vehicles (particularly drones).
A more detailed version of this text is available at linked-in
Applications shall be sent by e-mail to ansgar.radermacher at cea.fr
Applications have to contain detailed experience and background
information and should contain a recommendation letter.
Ansgar Radermacher CEA/DRT/DILS/LSEA
phone: +33 16908 3812
mailto: ansgar.radermacher at cea.fr
More information about the ecoop-info