Anelia Angelova

Google Brain

Title: Self-supervised Visual Learning for Robotics

Abstract: Recent advances in robotics bring the promise of ubiquitous robots much closer. Learning and adapting to the environment are required characteristics for robots deployed in real-life. But how can one provide adequate supervision for learning?

In this talk we will look into our recent work on self-supervised visual learning, which is learning from raw visual data and when supervision is provided by the robot itself. For example, the robot can learn about objects or the scene by moving in the environment, or can learn by using its own actions, additional sensors or constraints. Robots can also learn with minimal supervision, e.g. by observing passive third party imitations.

David Hogg

University of Leeds

Title: Laying the Perceptual Groundwork for Abstraction

Abstract: A fundamental question about human cognition is how we reason about assemblies of physical objects, and how we learn to do this. For example, when presented with an image of two interlocking rings, we see immediately that they can’t be separated. How is this possible, and where does the knowledge come from? Recent results from AI, including work on deep learning, suggest ways in which learning and reasoning abilities of this kind could be acquired from observing people handling assemblies of objects in the wild.

Elin Anna Topp

Lund University, http://cs.lth.se/elin_anna_topp/

Title: Skills for Industrial Robotic Assembly: Representation, Acquisition, Reuse and Adaptation

Abstract: Recently, a number of rather lightweight and inherently safe robots like the KUKA iiwa or ABB’s YuMi have entered the market. These robots are designed to work in close vicinity and even collaboration with human workers, e.g. in industrial assembly tasks. One issue in this context is to provide the robot with the necessary knowledge to handle its task and the interaction with its human co-worker. For companies working with such potential teams of robots and humans, it is not feasible to call in the otherwise necessary robot programming experts, if changes of the task description or other adaptations of the system have to be made. Even though the mentioned robots very often come with some form of basic skill libraries, these skills might not cover all the flexibility required in the targeted new production settings.

This talk will give an overview of a knowledge based approach to a skill representation for which we have during the recent years been able to provide and investigate suitable tools for user support for skill acquisition, reuse and adaptation. Skills in this case are seen as hierarchically organised sequences of actions and / or skills, i.e. they can be nested. Our approach also supports the acquisition and handling of dual-arm operations, which makes it particularly useful in the context of industrial assembly tasks.

Sinan Kalkan

Middle East Technical University, http://www.kovan.ceng.metu.edu.tr/~sinan/

Title: Context in robots

Abstract: In many of our cognitive capabilities and perceptual processes, we (humans) rely on vast amount of background knowledge and stimuli that we call context. Robots that are expected to populate our daily lives in near future should use such contextual information, and perceive, act, reason and plan accordingly. In this talk, after providing a short summary of how context helps us (humans) and the use of contextual cues in the robotics literature, I will present some of our recent work on incremental and hierarchical modeling of context in robots.

Karinne Ramirez-Amaro

Technische Universität München, http://www.ics.ei.tum.de/people/ramirez/

Title: Semantic Reasoning Method for the Understanding of Human Actions

Abstract: Autonomous robots are expected to learn new skills and to re-use past experiences in different situations as efficient, intuitive and reliable as possible. Robots need to adapt to different sources of information, for example, videos, robot sensors, virtual reality, etc. Then, to advance the research in the understanding of human movements, in robotics, the development of learning methods that adapt to different datasets are needed. In this talk, I will introduce a novel learning method that generates compact and general semantic models to infer human activities. This learning method allows robots to obtain and determine a higher-level understanding of a demonstrator’s behavior via semantic representations. First, the low-level information is extracted from the sensory data, then a meaningful semantic description, the high-level, is obtained by reasoning about the intended human behaviors. The introduced method has been assessed on different robots, e.g. the iCub, REEM-C, and TOMM, with different kinematic chains and dynamics. Furthermore, the robots use different perceptual modalities, under different constraints and in several scenarios ranging from making a sandwich to driving a car assessed on different domains (home-service and industrial scenarios). One important aspect of our approach is its scalability and adaptability toward new activities, which can be learned on-demand. Overall, the presented compact and flexible solutions are suitable to tackle complex and challenging problems for autonomous robots.

Yezhou Yang

Arizona State University, https://yezhouyang.engineering.asu.edu/

Title: Recognition Beyond Appearance, and its Robotic Applications

Abstract: The goal of Computer Vision, as coined by Marr, is to develop algorithms to answer What are Where at When from visual appearance. The speaker, among others, recognizes the importance of studying underlying entities and relations beyond visual appearance, following an Active Perception paradigm. The talk will present the speaker's efforts over the last several years, ranging from 1) hidden entities recognition (such as action fluent, human intention and force prediction from visual input), through 2) reasoning beyond appearance for solving image riddles and visual question answering, till 3) their applications in a Robotic visual learning framework as well as for Robotic Visual Search. The talk will also feature several ongoing projects and future directions among the Active Perception Group (APG) with the speaker at ASU School of Computing, Informatics, and Decision Systems Engineering (CIDSE).

Yu Sun

University of South Florida, http://www.cse.usf.edu/~yusun/

Title: Functional Object-Oriented Network (FOON) for Manipulation-knowledge Representation

Abstract: Objects exhibit their functionalities with their manipulation motions. Therefore, we connect functional-related objects and their relative motions (functional motions) during manipulation tasks to form a functional object-oriented network (FOON). This talk will present our approach of extracting FOON from online instructional videos and using FOON to give robots detailed manipulation instructions.