Dennis Herzog

Learning Actions from Observations

In the area of imitation learning, one of the important research problems is action representation. There has been a growing interest in expressing actions as a combination of meaningful subparts called action primitives. Action primitives could be thought of as elementary building blocks for action representation. In this article, we present a complete concept of learning action primitives to recognize and synthesize actions. One of the main novelties in this work is the detection of primitives in a unified framework, which takes into account objects and actions being applied to them. As the first major contribution, we propose an unsupervised learning approach for action primitives that make use of the human movements as well as object state changes. As the second major contribution, we propose using parametric hidden Markov models (PHMMs) for representing the discovered action primitives. PHMMs represent movement trajectories as a function of their desired effect on the object, and we will discuss 1) how these PHMMs can be trained in an unsupervised manner, 2) how they can be used for synthesizing movements to achieve a desired effect, and 3) how they can be used to recognize an action primitive and the effect from an observed acting human.

Parametric Hidden Markov Models for Recognition and Synthesis of Movements

Disclosed is a method for controlling the flame front during the in situ combustion of a subterranean carbonaceous stratum which involves monitoring the extent and movement of said flame front to determine the location of one or more segments of the flame front which exhibit unfavorable combustion characteristics, and injecting one or more gases into the vicinity of one or more of said segments to control and optimize the combustion in said segment.

Motion imitation and recognition using parametric hidden Markov models

The recognition and synthesis of parametric movements play an important role in human-robot interaction. To understand the whole purpose of an arm movement of a human agent, both its recognition (e. g., pointing or reaching) as well as its parameterization (i. e., where the agent is pointing at) are important. Only together they convey the whole meaning of an action. Similarly, to imitate a movement, the robot needs to select the proper action and parameterize it, e. g., by the relative position of the object that needs to be grasped. We propose to utilize parametric hidden Markov models (PHMMs), which extend the classical HMMs by introducing a joint parameterization of the observation densities, to simultaneously solve the problems of action recognition, parameterization of the observed actions, and action synthesis. The proposed approach was fully implemented on a humanoid robot HOAP-3. To evaluate the approach, we focused on reaching and pointing actions. Even though the movements are very similar in appearance, our approach is able to distinguish the two movement types and discover the parameterization, and is thus enabling both, action recognition and action synthesis. Through parameterization we ensure that the synthesized movements can be applied to different configurations of the external world and are thus suitable for actions that involve the manipulation of objects.

Intuitive Skill-Level Programming of Industrial Handling Tasks on a Mobile Manipulator

In order for manufacturing companies to remain competitive while also offering a high degree of customization for the customers, flexible robots that can be rapidly reprogrammed to new tasks need to be applied in the factories. In this paper we propose a method for the intuitive programming of an industrial mobile robot by combining robot skills, a graphical user interface and human gesture recognition. We give a brief introduction to robot skills as we envision them for intuitive programming, and how they are used in the robot system. We then describe the tracking and gesture recognition, and how the instructor uses the method for programming. We have verified our approach through experiments on several subjects, showing that the system is generally easy to use even for inexperienced users. Furthermore, the programming time required to program a new task is very short, especially keeping traditional industrial robot programming methods in mind.

Recognition and Synthesis of Human Movements by Parametric HMMs

The representation of human movements for recognition and synthesis is important in many application fields such as: surveillance, human-computer interaction, motion capture, and humanoid robots. Hidden Markov models (HMMs) are a common statistical framework in this context, since they are generative and are able to deal with the intrinsic dynamic variation of movements performed by humans. In this work we argue that many human movements are parametric, i.e., a parametric variation of the movements in dependence of, e.g., a position a person is pointing at. The parameter is part of the semantic of a movement. And while classic HMMs treat them as noise, we will use parametric HMMs (PHMMs) [6,19] to model the parametric variability of human movements explicitly. In this work, we discuss both types of PHMMs, as introduced in [6] and [19], and we will focus our considerations on the recognition and synthesis of human arm movements. Furthermore, we will show in various experiments the use of PHMMs for the control of a humanoid robot by synthesizing movements for relocating objects at arbitrary positions. In vision-based interaction experiments, PHMM are used for the recognition of pointing movements, where the recognized parameterization conveys to a robot the important information which object to relocate and where to put it. Finally, we evaluate the accuracy of recognition and synthesis for pointing and grasping arm movements and discuss that the precision of the synthesis is within the natural uncertainty of human movements.

Communicating Unknown Objects to Robots through Pointing Gestures

Delegating tasks from a human to a robot needs an efficient and easy-to-use communication pipeline between them - especially when inexperienced users are involved. This work presents a robotic system that is able to bridge this communication gap by exploiting 3D sensing for gesture recognition and real-time object segmentation. We visually extract an unknown object indicated by a human through a pointing gesture and thereby communicating the object of interest to the robot which can be used to perform a certain task. The robot uses RGB-D sensors to observe the human and find the 3D point indicated by the pointing gesture. This point is used to initialize a fixation-based, fast object segmentation algorithm, inferring thus the outline of the whole object. A series of experiments with different objects and pointing gestures show that both the recognition of the gesture, the extraction of the pointing direction in 3D, and the object segmentation perform robustly. The discussed system can provide the first step towards more complex tasks, such as object recognition, grasping or learning by demonstration with obvious value in both industrial and domestic settings.