Jan Peters
German Research Center for AI (DFKI), Research Department: SAIROL
Institute for Intelligent Autonomous Systems
TU Darmstadt
Title of the talk: Inductive Biases for Robot Reinforcement Learning
Abstract:
Autonomous robots that can assist humans in situations of daily life have been a long standing vision of robotics, artificial intelligence, and cognitive sciences. A first step towards this goal is to create robots that can learn tasks triggered by environmental context or higher level instruction. However, learning techniques have yet to live up to this promise as only few methods manage to scale to high-dimensional manipulator or humanoid robots. In this talk, we investigate a general framework suitable for learning motor skills in robotics which is based on the principles behind many analytical robotics approaches. To accomplish robot reinforcement learning learning from just few trials, the learning system can no longer explore all learn-able solutions but has to prioritize one solution over others – independent of the observed data. Such prioritization requires explicit or implicit assumptions, often called ‘induction biases’ in machine learning. Extrapolation to new robot learning tasks requires induction biases deeply rooted in general principles and domain knowledge from robotics, physics and control. Empirical evaluations on a several robot systems illustrate the effectiveness and applicability to learning control on an anthropomorphic robot arm. These robot motor skills range from toy examples (e.g., paddling a ball, ball-in-a-cup) to playing robot table tennis, juggling and manipulation of various objects.
Martin Adams
Department of Electrical Engineering
Universidad de Chile
Title of the talk: Advancing State Estimation: Insights from Random Finite Sets
Abstract:
Autonomous navigation, mapping, and multi-target tracking are state estimation problems
relevant to a wide range of applications. In such applications, estimation has very little meaning
without a clear concept of estimation error. State-of-the-art solutions have traditionally been
formulated using random vectors (RVs) in stochastic filtering, smoothing, or optimization based
approaches, but fail to jointly minimize both spatial and detection errors, without the use of addon heuristics. In the Simultaneous Localization and Mapping (SLAM) problem, this usually takes
the form of a back end solver and an independent front end, necessary for implementing the
detection and measurement-to-state association heuristics. In contrast to RV-based approaches,
the use of Random Finite Sets (RFS) yields a general measurement likelihood, which allows data
association and map management routines to be a native part of the entire estimation approach,
effectively combining the SLAM back and front ends into a single, joint estimation framework.
RFS formulations also allow the joint consideration of detection and clutter statistics within the
estimator and have recently attracted considerable research interest as well as deployment in
commercial applications.
As well as justifying the general application of RFS frameworks, this presentation will focus on
new RFS-based solutions to SLAM. To encompass the advantages of recent Maximum likelihood
(ML) batch approaches, which use sparse matrix methods such as the g2o solver, it will be shown
that the SLAM state can be modelled as a mixed distribution. This distribution jointly represents
the vector-valued trajectory and the RFS-valued map, and is referred to as the Vector-Generalized
Labeled Multi Bernoulli (V-GLMB) distribution. This yields hybrid RFS-RV-SLAM solutions,
which yield competitive, and often superior, results to their RV-SLAM counterparts, while
circumventing the need for fragile data association methods. A framework for solutions which
combine the SLAM back and front ends into a single ML estimation framework will be
demonstrated.
Darius Burschka
Robotics, Artificial Intelligence and Embedded Systems
Technical University of Munich
Title of the talk: Robust and Efficient Coupling of Perception to Actuation in Dynamic Environments
Abstract:
I will discuss the problem of efficient and robust information exchange
between the perception and actuation modules in manipulation and mobile
systems. While most current approaches use three-dimensional
representations of the world as an interface to generate actions in a
robot, this is not the native representation neither of the sensor nor
of the motion controller and it requires calibration parameters to
calculate. It is prone to errors and drifts making the accuracy of the
system unreliable. Direct definitions of the tasks in the sensor space
or some lower dimensional abstraction allows a robust operations of the
systems in dynamic environments.
I will show examples of non-metric task representation for navigation,
path planning, obstacle avoidance and manipulation.
Paulo Drews-Jr
Center for Computational Sciences
Federal University of Rio Grande
Title of the talk: Robotics and its Challenges in the World of Artificial Intelligence
Abstract:
Robotics brings important challenges from both a scientific and technological point of view. The relevance and practical applications make efforts necessary. Within an extremely interdisciplinary context of robotics, autonomy brings important and relevant challenges and opportunities that converse with recent advances in artificial intelligence. Practical examples developed in projects by the Automation and Intelligent Robotics group – NAUTEC and the Embrapii Center for Robotics and Artificial Intelligence – iTEC, both from the Federal University of Rio Grande – FURG, Brazil, will also be discussed. These robotics examples includes aerial, underwater, hybrid aerial-underwater, logistics, agriculture, among other applications.
Jorge Solis
Faculty of Health, Science and Technology
Department of Engineering and Physics
Karlstad University
Title of the talk: Challenges towards Industry 5.0 in production systems: from collaborative robots to energy efficiency
Abstract:
Today and tomorrow's industry requires modern technology where collaboration takes place between people, between people and machines and between machines. Machines can monitor themselves, analyze the results and autonomously optimize operating conditions and production. The result is higher efficiency and productivity. In this lecture, I will present an overview of the ongoing research projects at Karlstad University within the applications areas to ageing, energy, environment and education. In particular, the challenges within Industry 5.0 will be exemplified with some ongoing research projects that deals with a sustainable human-robot synergy in assembling tasks as well as adaptive battery energy storage in the food industry. Some possible extensions and potential applications will be outlined.
