Report generation from episodic robot memories serves two purposes:

  * Getting an overview of what an autonomous robot has done, which plans it executed, which parameters led to its decisions,
  * Forming a comprehensive overview of what happened during an experiment at a glance, and
  * Helping developers detecting (subtle) anomalies in robot behaviour early

The memories from which reports are to be generated are quite big and not easy to comprehend when looking at the raw data. The memories (or logs) are well structured in OWL (XML, basically) and reflect the structure of a hierarchical tree. You can see an example in this file:

  * http://www.robots-doing-things.com/incoming/log-packaged-2014-12-30-16-10-37.tar.gz (see file `cram_log.owl`)

This is just an example log, and the ones from which reports are to be generated will have different content. The structure, though, will stay the same.

These logs are generated using a component called `Semantic Hierarchical Recorder`, or `semrec` for short. You can find this open source tool (written and maintained by our institute) here:

  * https://github.com/code-iai/semrec

The components to be developed during the GSoC project should suffice the following (yet very rough) guidelines:

  * Mainly be written in Java
  * Integrated into an existing framework that mainly uses (SWI) Prolog that can call the Java code you developed

Once it is complete, the final component will be used in the OpenEASE robot memory analysis portal:

  * http://www.open-ease.org/

If you haven't done any Prolog yet, doing a tutorial or two before the project is advisable (to get the grips of how to write simple predicates). If you haven't developed any Java code yet, and haven't done any C++ either (because those are very similar), you might start off to a bumpy ride. Since the final component will require analysing XML code through Java (and parsing some JSON as well), you will need some good understanding of what is going on in Java. Other than that, no `special` requirements must be met.

A plus definitely is some understanding in statistics. Doing robot experiments produces all kinds of data (symbolic and subsymbolic), and includes a lot of task durations, trajectory lengths, and so forth. The final report that is generated from logs would ideally include confidence intervals, significance levels of parameters, and variances. If you haven't ever heard of any of those, please look them up - if you have, and you liked the statistics 101 in your school, that's great and helps a lot.

All of the above describe an ideal case - just ask if anything is unclear.