Adaptive Safety
in Autonomous Ecosystems
Dávid Halász
halasz@mail.muni.cz
What are (software) ecosystems?
● An evolutionary step towards more advanced software systems
● Systems of Systems
● Higher complexity
● Higher uncertainty
Source: roomsketcher.com
Source: shutterstock.com
Autonomous Ecosystems
● Higher degree of autonomy
● Member systems can join or leave at any time
● Increased dynamicity of context changes
● Lower amount of predictable situations
● Harder to achieve safe and secure behavior
● Traditional safety mechanisms are not enough
Problems with safety
● Intentional vs unintentional behavior
● False positive and false negative situations
● Supervision awareness
● Feedback loops
Related work
● Simplex Architecture
● Isolated Environments
● Autonomous Vehicles (Danish)
● Wireless Networks (Bacem)
● MAPE-K
● Runtime models
Proposed solution
Trust-based Adaptive Safety
● Trust quantification method
○ Trust Team is working on it
○ Overlap with Danish’s research
○ Needs to have a more granular (non binary) output
● Real-time trust evaluated among member systems
○ Each against each
○ Direct and indirect trust
● Dynamic toggling of features and safety mechanisms
○ This is where it is adaptive
Expected output
● A method for adaptive safety
● A software architecture
● Validation using simulations
● Reaching out to automotive companies
● Focus is on AVs, but can be generalized
Thank you for your attention!
Dávid Halász
halasz@mail.muni.cz