Artificial Intelligence Platform (PFIA) 2015
Jul 24, 2015 |
👴 This article is about a discontinued version of craft ai
, some information may not reflect what craft ai
can currently do!
For more up to date reading, check our blog
and do not hesitate to visit our website
Early july was held the “plateforme intelligence artificielle (PFIA)” (artificial intelligent platform). This event is a week of interdisciplinary, and mostly academic, AI conferences.
Some of us were involved in this event for different purposes: as program committee member, as round-table contributor, as author, speaker and also as simple attendee.
During the week of the event, the craft ai team was in tune with innovation and novelties presented by the national academic and industrial research on AI. The topics were practical applications of AI, planning/learning/decision, fundamental AI, knowledge engineering and multi-agent systems.
What we did
We were mostly involved in three events at PFIA: the game and AI workshop, the application of artificial intelligence conference (APIA) and the RJCIA conference dedicated to new researchers. During this events, we presented especially our works on AI related to video games.
At RJCIA and during the game and AI workshop, Juliette Lemaître, PhD student at craft ai/MASA Group and the Heudiasyc lab (UMR UTC/CNRS) presented her work on designing strategy to control opponent AI. Most of strategic games (especially 4X games like Civilization, Endless Legend, Master of Orion, etc.) are based on resource management, but you often have to balance concurrent domains as military, culture, religion… The vast world and numerous possibilities make it difficult to give control to game designers to allow them to obtain the desired behavior. This work is about giving them the right tools, intelligible and transparent.
At the APIA conference, Caroline Chopinaud presented the OCTAVIA research project that focuses on the development of a tool dedicated to game designers to provide a virtual environment with a rich and credible ambiance. Ambiance partly exists through the execution of credible and interactive behaviors by the non-player characters that evolve in a virtual environment. Our main goal is to help designers to model such an ambiance and to quickly breathe life into empty environments. Craft ai automates the orchestration of the NPC’s behaviors involved in scenes to create a more global ambiance. The work we done on this project was also published in Game AI Pro 2.
During the AI&Games workshop, we participated to a round-table about how game AI competition should be held. Researcher are interested in trying their algorithm on challenging problems proposed by industrial and instantiated in a video game. The aim of this competition is to tightened the links between industry and academia and to create a synergy between these two worlds. To put this kind of competition in place, many issues have been discussed during the round-table such as the means to evaluate an AI in a game. Based on optimality? reusability? fun? cpu time? human like? Maybe a mix between some of these criteria. The discussion was focused on the pros and cons of this type of competition for industry and academia:
|providing an API costs (extra) money
||fun (especially for students)
hard to publish on those kind of competition
What we saw : Invited talk with Rodney Brooks
There were of course many other conferences where we were not directly involved with very interesting topics. This talk was one of them. Rodney Brooks (founder of iRobot and former director of the MIT Computer Science and AI Lab) talked about his experience in robotics.
Here’s a little game: can you find the 4 mistakes hidden in this roomba (first version)? According to its creator 4 buttons, 4 mistakes! What does 0/1 mean to turn it on/off? Small/Medium/Large button room selection : but what is a medium sized room? Quite funny indeed.
Another nice point was his view on the price of robots. Here he claimed that today’s cheapest robot is fully autonomous (say the roomba) and the most expensive are human controlled. It is of course purely subjective, but still, it got me thinking.
The price of the roomba was fixed before its design, forcing any design choice to respect the price constraint. It has been chosen after asking people “What is the maximum price of an object you can buy without asking your spouse ?” (around 200 euros).
Baxter is an industrial but human safe robot. It can be programmed without any technical skills, and share the work with a human. Tip of the day: Baxter signals its intention to humans by looking at the object it wants to interact with, thus nearby humans are not surprised when the robot starts to move.
Rodney Brooks concluded his talk by giving his vision on the state of the art on the three aspects that limit domestic (mobile) robotics: Mobility (not ok), Manipulation (not ok), Messiness (ok, thanks to the kinect).
Things I liked
There were also lots of interesting talks that I would like to mention, that might be presented in a later article:
The work of Nicolas Seydoux, et al. on Semantics and IoT
The invited talk of Hector Geffner (ICREA & Universitat Pompeu Fabra, Barcelona, SPAIN), especially the presentation of the Iterated Width algorithm (IW)
The project ethicaa working on the ethic of virtual agents.
PFIA was also the opportunity to talk directly with many people, to meet friends. This conference really succeeded in providing a nice place to exchange ideas. We are looking forward the next event (but we’ll have to wait until 2017).