Towards adaptive trajectories for mixed autonomous and human-operated ships

We are witnessing the rise of autonomous cars, which will likely revolutionize the way we travel. Arguably, the maritime domain lags behind, as ships operate on many more degrees of freedom (thus, a much larger search space): there is less physical infrastructure, and rules are less consistent and constraining than what is found on roads. The problem is further complicated by the inevitable co-existence of autonomous and human-operated ships: the latter may take unpredictable decisions, which require adjustments on the autonomous ones. Finally, the problem is inherently decentralised, there is no central authority, and communication means can be very diverse in terms of communication distance and performance, mandating special care on which information is shared and how. In this work, we elaborate on the challenges of trajectory prediction and adaptation for mixed autonomous and human-operated ships, and we propose initial ideas on potential approaches to address them.

1. Towards adaptive trajectories
for mixed autonomous and
human-operated ships
Danilo Pianini, Sven Tomforde

SISSY Workshop, Aarhus, 24/16/09

2. Navigation for Autonomous Ships
Maritime Autonomous Surface Vessels (MASS)
-First systems operating
-Several research initiatives with a focus on local conditions and restrictions
-Developing market, lots of efforts

One of the major problems: Navigation of MASS
-Long-term path planning (seamaps, ship and obstacle information, conditions)
-Short term behaviour (collision avoidance, response to dynamic objects)

3. Scenario: The Kiel Fjord
Kiel, Germany
-250.000 inhabitants, Northern Germany
-270 km south from Aarhus
-Divided by water
-Kiel Fjord is the mouth of the Kiel Canal
-Fjord and canal are among the busiest waterways in Europe
Public transport
-Partly done via ferries
-Bay of Kiel
Aarhus
Kiel

4. Scenario: The Kiel Fjord (2)
Vision: Autonomous Ferries
-“MS Wavelab”: Platform for R&D
-Increasing level of autonomy
Shipping @Kiel as SISSY problem
-Variety of autonomous entities
-All are part of the system “Fjord”,
incl. rules and conditions, use the
same technology (e.g. AIS)
-Decision about integration: information sharing, coordination, etc.

5. The problem
Autonomous navigation
-Depends on local perception
-May integrate external perception
-Basis: Communication
Communication
-Changing availability of modes
-Varying characteristics
(e.g. bandwidth, range)

6. The problem (2)
Adaptive communication as a basis for autonomous navigation
-Goal: Opportunistic utilisation of available modes
-Continual modification of the available communication means and their performance
Challenges:
-Heterogeneity
-Mixed autonomic and human operation
-Partial information
-Communication challenges
-adapt data to the channel
-opportunistically select the most appropriate challenge for the recipient
-Safety and security

7. A different approach through aggregate computing
Classic approach: define communication protocols, let autonomous entities adhere
-Explicitly-designed interaction
-Very reasonable in all cases in which a clear and immutable protocol can be established
-Ill-suited for the specific problem at hand: very complex protocol design
-Hardly scalable
- No reuse, composition, or extension mechanisms
Aggregate computing: program all entities as a single computational machine. The compiler/interpreter figures out the communication protocol
-Interaction is implicit in the macroscopic program
-Programs manipulate distributed data structures called fields
-Extremely scalable
- The same reuse, composition, and extension abstractions of a functional language
- Changing the program automatically updates the interaction protocol

8. Field-based approaches to trajectory computation
A natural application of aggregate computing:
-shortest path
-obstacle avoidance
-anticipatory behavior
Open issues:
-heterogeneity: consider ship
capabilities and conditions
(e.g., velocity) in anticipatory
adaptation
-integrate learning as part of the
anticipatory predictor

9. Adaptive communication through anisotropic fields
The XC Calculus however introduced anisotropic fields (through a
new exchange primitive): their value in a point depends on the
observer.
-Promising for fields whose data changes depending on the
communication channel capacity
-Unknown impact on the API: exchange is a very low-level
operator
-can we adapt the existing libraries?
Communication in classic aggregate programming is
isotropic: values are uniform regardless of the direction they
are been looked from.

10. Information dependability through error-aware fields
Idea: fields may embed a notion of error
-the core idea is to let ships know that the information
is partial, incomplete, or only partially reliable
-and how reliable, in a quantifiable way
-error accumulates with hops
-error is greater when a channel doesn’t allow for all
data to be sent through
-error varies with time
-complements or replaces anisotropic fields
-may be implemented on top of anisotropic fields?
-different accumulated error makes anisotropic fields
redundant?
-similar to anisotropic fields, the impact on the
high-level API is unknown at the moment

11. Runtime reconfiguration via pulverization
Aggregate computing assumes a certain degree of homogeneity:
-enough computational power in every node
-similar sensors and actuators
-capability to host state
-communication means
In our contest, these are not guaranteed! Devices are extremely heterogeneous
-a small sailing boat is very different than a large cruiser or a cargo ship under all the previous points of view
Pulverization
<- split the computation into sub-components
-the division happens at the framework level, the program is written as if the
split never occurred
-aggregate computing is “naturally pulverizable” in at least 5 components
-in an homogeneous system, all components could be executed on any device
deploy the sub-components arbitrarily ->
-some components can be constrained, e.g., sensors must be bound to devices
with physical sensors
-The program logic is unaware of that is executing on heterogeneous devices
P2P, homogeneous
Cloud instance
edge server
sensoactuators

12. Conclusion
Ships as complex self-integrating systems
-Navigation depends on dynamic perception and static knowledge
-Communication among ships to share information
-Aggregate computing model to program all participants as a single
computational machine