How to Reduce Computation Time While
Sparing Performance During Robot
Navigation? A Neuro-Inspired
Architecture for Autonomous Shifting
Between Model-Based and Model-Free
Learning
R´ emi Dromnelle
1(B) , Erwan Renaudo
2 , Guillaume Pourcel
1 ,
Raja Chatila
1 , Benoˆ ıt Girard
1 , and Mehdi Khamassi
1
1 Institut des Syst` emes Intelligents et de Robotique (ISIR), Sorbonne Universit´ es,
CNRS, 75005 Paris, France
remi.dromnelle@gmail.com
2 Intelligent and Interactive Systems Lab (IIS), Universit¨ at Innsbruck,
6010 Innsbruck, Austria
Abstract. Taking inspiration from how the brain coordinates multiple learning systems is an appealing strategy to endow robots with
more flexibility. One of the expected advantages would be for robots
to autonomously switch to the least costly system when its performance
is satisfying. However, to our knowledge no study on a real robot has
yet shown that the measured computational cost is reduced while performance is maintained with such brain-inspired algorithms. We present
navigation experiments involving paths of different lengths to the goal,
dead-end, and non-stationarity (i.e., change in goal location and apparition of obstacles). We present a novel arbitration mechanism between
learning systems that explicitly measures performance and cost. We find
that the robot can adapt to environment changes by switching between
learning systems so as to maintain a high performance. Moreover, when
the task is stable, the robot also autonomously shifts to the least costly
system, which leads to a drastic reduction in computation cost while
keeping a high performance. Overall, these results illustrates the interest
of using multiple learning systems.
1 Introduction
The idea of taking inspiration from how the brain coordinates multiple learning
systems to enable more flexibility in robots is getting more and more attention
in the robotics community [1–6]. One of the expected advantages of such a
strategy would be for robots to autonomously learn which system is the most
appropriate for each encountered task or situation. For instance, a robot can
learn that different systems are efficient in different subparts of the environment
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 68–79, 2020.
https://doi.org/10.1007/978-3-030-64313-3_8
Sparing Performance During Robot
Navigation? A Neuro-Inspired
Architecture for Autonomous Shifting
Between Model-Based and Model-Free
Learning
R´ emi Dromnelle
1(B) , Erwan Renaudo
2 , Guillaume Pourcel
1 ,
Raja Chatila
1 , Benoˆ ıt Girard
1 , and Mehdi Khamassi
1
1 Institut des Syst` emes Intelligents et de Robotique (ISIR), Sorbonne Universit´ es,
CNRS, 75005 Paris, France
remi.dromnelle@gmail.com
2 Intelligent and Interactive Systems Lab (IIS), Universit¨ at Innsbruck,
6010 Innsbruck, Austria
Abstract. Taking inspiration from how the brain coordinates multiple learning systems is an appealing strategy to endow robots with
more flexibility. One of the expected advantages would be for robots
to autonomously switch to the least costly system when its performance
is satisfying. However, to our knowledge no study on a real robot has
yet shown that the measured computational cost is reduced while performance is maintained with such brain-inspired algorithms. We present
navigation experiments involving paths of different lengths to the goal,
dead-end, and non-stationarity (i.e., change in goal location and apparition of obstacles). We present a novel arbitration mechanism between
learning systems that explicitly measures performance and cost. We find
that the robot can adapt to environment changes by switching between
learning systems so as to maintain a high performance. Moreover, when
the task is stable, the robot also autonomously shifts to the least costly
system, which leads to a drastic reduction in computation cost while
keeping a high performance. Overall, these results illustrates the interest
of using multiple learning systems.
1 Introduction
The idea of taking inspiration from how the brain coordinates multiple learning
systems to enable more flexibility in robots is getting more and more attention
in the robotics community [1–6]. One of the expected advantages of such a
strategy would be for robots to autonomously learn which system is the most
appropriate for each encountered task or situation. For instance, a robot can
learn that different systems are efficient in different subparts of the environment
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 68–79, 2020.
https://doi.org/10.1007/978-3-030-64313-3_8