Community Research and Development Information Service - CORDIS


Phoenix Report Summary

Project ID: 665347
Funded under: H2020-EU.1.2.1.

Periodic Reporting for period 1 - Phoenix (Exploring the Unknown through Reincarnation and Co-evolution)

Reporting period: 2015-10-01 to 2016-09-30

Summary of the context and overall objectives of the project

"What is the issue being addressed?
Phoenix is a method to provide access to environments that are unknown and inaccessible. This includes any kind of environment, manmade or natural, be it a mixing tank full of toxic chemicals, the depths below the Fukushima reactor or, one day, the sea of a distant planet.

Why is it important for society?
Since the dawn of humankind, humans have explored the Earth directly or through measurement devices. Countless discoveries arose from that activity. Clearly the value of exploration cannot be underestimated, however there are some specific use-cases where direct (human) exploration is not possible, such as:
– Mapping of a pipeline to find obstructions, leaks or faults. This could be part of a strategy to more efficiently deliver drinking water or to prevent contamination.
– Exploring underground channels, which cannot be otherwise accessed without damaging them. This could be part of a strategy related to more efficient extraction of oil or natural gas or to the search of natural CO2 storage locations.
– Measuring from the depths of glaciers or inside volcanos. This could be part of a strategy to better model climate change.
At the moment we focus on the first use-case. We believe that the outcome from this line of research will be part of the foundation we need to elaborate more complex use-cases like the second and third one above.

What is the overall objective?
The objective of the Phoenix is to develop a method to explore unknown and inaccessible environments. This involves: 1) the development of a co-evolutionary framework (as described in the proposal text), 2) the design of versatile agent technology and 3) techniques to formalize different kinds of expert knowledge, even uncertain knowledge to influence the design of agents and the evolutionary algorithm which influences their evolution and "rebirth" in the Phoenix system.
Achieving this goal will also shed light on emergent properties of self-organization, local adaptation and division of labour in autonomous systems."

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

"During the 1st year of the project a number of the challenges associated with the development of the agents have been worked on. The agents have the task to explore the unknown and otherwise inaccessible environment and monitor and record a variety of sensors during this exploration for later retrieval. The two major bottlenecks for this task are 1) the amount of energy consumed for data capturing and data storage, and 2) the amount of memory required to store all the data. As the physical agents are only cm3 in size, yet should be able to measure during a day or more, energy and storage are extremely scarce resources. Addressing this issue requires "smart data compression", i.e. the efficient collection of sensor data. To those ends, a careful analysis of the trade-off between compression performance (affecting the agent’s memory consumption), and its hardware implementation (affecting the agent’s energy consumption) has been performed for a wide range of data selection and data compression techniques.

The use of physical agents (i.e. motes) to explore unknown and (near-)inaccessible environments involves expertise. Artificial evolution could ensure that agents evolve to explore unknown and inaccessible environments efficiently. However, understanding the goal and constraints of the exploration, and providing an interpretation of the data acquired to address that goal requires “high level” reasoning. Accordingly, we utilize a knowledge representation scheme to formally represent and reason on exploration procedures. We develop a knowledge base to store the knowledge needed in these exploration tasks. The knowledge base includes different kinds of knowledge (including domain and procedural knowledge) elicited from experts and from textual sources, which is represented in a machine-readable, extendible format. It is able to take represent and reason on different kinds of knowledge with varying degrees of certainty.

To these ends, informed by the results of a literature review on the state-of-the-art, we developed an ontological representation that allows us to represent and reason on uncertain knowledge using uncertain graphs. At the time of writing, we are researching methods to extend our scheme to represent uncertain procedural knowledge. Additionally, in accordance with the planning for this work package, we are also beginning to plan for the Human Interface Layer as described in the proposal text.

Finally, we have implemented a first version of an evolutionary algorithm that is able to detect the intrinsic features of an unknown optimization problem and, by using the available knowledge from the existing literature in the Evolutionary Computing domain (that we represented, for the first time, in an original ontology named ECO - Evolutionary Computing Ontology), adjust its parameters and operators to obtain superior optimization performance. This evolutionary algorithm is currently being tested on numerical benchmarks and simple simulation scenarios and will be extended in the continuation of the project to include the co-evolution of evolvable agents and environment models. In addition to that, we performed a comparative analysis of the existing fitness-free evolutionary methods to assess their performance in a simple exploration (maze navigation) task. This analysis represents a useful starting point for comparing fitness-based vs fitness-free evolutionary methods."

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

On the agent side, state-of-the-art research has focused on maximizing data compression with compute-intensive algorithms on one hand, and on lightweight compression schemes which minimize energy spent in the compression operation on the other hand. Important for the extremely energy-scarce Phoenix motes is however the energy consumption of the complete sensor node, including energy for data processing and compression (favoring lightweight schemes), as well as energy to write the compressed data to non-volatile memory (favoring maximal compression schemes). Hence, a complete energy model of the Phoenix mote, including its storage medium and various compression candidates has been made, to carefully assess this energy-vs-compression rate trade-off. This assessment has been done with Phoenix use-case representative waveforms.

Secondly, innovation has been brought towards adaptive information extraction and lossy information compression techniques, requiring little flexibility overhead. This is important in the Phoenix context, which strives towards configuring generations of versatile agents in terms of their data gather capabilities (more or less accurate data collection), and energy usage. To this end, various versatility knobs have been studies in terms of their system level energy and performance impact.

We also provided a knowledge representation scheme that is open and extendible (allowing the scheme to remain current as the state-of-the-art in technology, environmental modeling and in other domains advances). This scheme is highly novel, and will be capable of representing and reasoning on different kinds of knowledge with different degrees of (un)certainty.

The ability to reason on uncertain knowledge about unknown environments will form part of a meaningful contribution to increasing the efficiency of artificially evolving hardware, like that in Phoenix, and also to A.I., where related topics are at the cutting-edge of that field. In the year ahead, we foresee the opportunity to demonstrate the impact of this research with 1) a joint demonstration project, and 2) the development a human interface layer for Phoenix.

The proposed combination of a self-adaptive Evolutionary Algorithm and a knowledge-based approach goes beyond the state-of-the-art in that it represents a first step towards a fully autonomous optimization system that is capable to learn by experience, update its knowledge base, and adapt its rules, as opposed to currently existing methods based on predetermined (pre-programmed) ad hoc adaptation rules. Such a system may be applied in a number of optimization tasks beyond Phoenix, including –but not limited to– industrial design, network systems and logistics.

Related information

Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top