Informatica e Robotica per l

1
Informatica e Robotica per lAutomazione (IRA)
Lab.

• Informatica e Robotica per lAutomazione (IRA)
  Lab.
• Dipartimento Informatica, Sistemistica e
  Comunicazione,
• Università degli Studi di Milano Bicocca
• Domenico G. Sorrenti
• domenico.sorrenti_at_unimib.it
• http//www.disco.unimib.it/sorrenti
• lab. website http//www.ira.disco.unimib.it

2
Research activities

• In the IRAlab we pursue research in autonomous
  systems
• Most of our work concerns autonomous robots
• What are autonomous robots? Lets see 3 examples.

3
Autonomous robots - example 1
4
Autonomous robots - example 2
5
Autonomous robots - example 3
6
Autonomous robots - example 4
7
Autonomous robots - example 4
8
So, what are autonomous robots?

• A machine that receives information from its
  working environment by means of its sensors, and
  basing on this information decides autonomously
  what action to execute, out of the ones
  available, in order to pursue a given task.

9
Autonomous robots as controllers
complex modeling of physics control
systems programming
modeling of physics probabilistic
reasoning programming analog and digital
electronics
10
Staff

• Domenico G. Sorrenti, assoc. prof.
• Daniele Marzorati, research assoc.
• Axel Furlan, research assist.
• students

11
Main research topics

• Vision-based SLAM (Simultaneous Localization and
  Mapping)
• Multi-target tracking
• Benchmarking, a cultural approach, orthogonal to
  the topics.

12
Vision-based SLAM
costruzione di un modello (mappa) dellambiente
di lavoro
13
Vision-based SLAM

• we tackled the uncertainty modeling problem,
  relevant in vision-based SLAM
• we are now at the frontier of the research, at
  the international level, with the inverse
  scaling parameterization
• expectations
• to consolidate the results publishing comparisons
  with other proposals on known benchmarks
• to handle large maps (CISLAM approach from
  Zaragoza)
• to handle dynamic scenes
• this is a joint work with Politecnico di Milano
• cooperation Universidad de Zaragoza, Albert
  Ludwig Universiteit Freiburg

14
Multi-target tracking
15
Multi-target tracking

• we tackled the problem of handling ambiguities in
  tracking, relevant in vision-based navigation
  (indoor and outdoor)
• in this field we are now behind the frontier of
  the research, at the international level
• expectations
• to publish our current results in papers based on
  comparisons with other proposals, on known
  benchmarks
• to integrate with the research in SLAM o increase
  accuracy in both systems
• to apply our systems to both indoor (project
  Grandi Attrezzature 2008) and outdoor (robotic
  cars/buses) navigation.
• cooperation Politecnico di Milano

16
Benchmarking in SLAM
17
Benchmarking in SLAM

• we tackled the problem of procuring datasets for
  benchmarking of SLAM algorithm by any research
  groups, in a project funded by the EEC, quite
  successful
• this is a support of the research activity,
  though innovation had to be brought in for
• the definition of the Ground Truth collection
  systems for the robot pose,
• the definition of performance measures,
• the definition of some algorithms, i.e.,
  benchmark solutions
• in this field we are the frontier of the
  research, at the international level
• expectations
• to publish our results in papers devoted to
  robotic benchmarking
• to apply our SLAM research results to the
  collected datasets
• to maintain the lead in the field, e.g., by using
  the Grandi Attrezzature 2008 infrastructure for
  benchmarking research fields where offline
  datasets are not viable.
• cooperation Politecnico di Milano, Universidad
  de Zaragoza, Albert Ludwig Universiteit Freiburg

18
Why robotics is different?

• Taxonomy of research activity
• at the conceptual level pure pencil and paper
  work
• at the software implementation level software
  development
• at the hardware level robotics, and other real
  research activities.
• Real example
• during the SantAmbrogio week in December 2008
  about 10 people from UNIMIB and POLIMI spent the
  something more than one week, after preparing for
  this activity, collecting indoor datasets for
  SLAM benchmarking in the Bicocca location.
• about 8 top-qualified people from UNIZR and ALUFR
  spent at about the same amount for checking and
  validating the datasets.
• End of January it turned out that a bug in a
  newer version of the IEE1394 (Firewire) driver
  invalidated all such datasets.
• The datasets had to be re-collected it took
  about one month of full pipelined work for the
  same number of people, which amounts to more than
  one year of labor.
• Pure pencil and paper research do not risk such
  troubles, and the same applies to pure software
  research.

19
Conclusions

• Both SLAM and multi-target tracking are required
  for autonomous driving, in any application
  robotized car/buses, indoor autonomous navigation
  services, unmanned ground / air / underwater
  vehicles.
• In a few years many robotic product will reach
  the market shall we be ready for selling them or
  we will be just customers?