Stefan Schaal - US grants

Even apparently simple tasks like reaching for an object involve the simultaneous graded activation of a large number of joints and muscles. This project investigates the principles that underlie the generation of such multi-joint movements, and how perceptual information guides these actions. Whereas many of the current theoretical approaches are limited to either rhythmic or discrete movements, this project will develop an integrated framework that encompasses both movement types and therefore can be applied to complex movements. The investigation will proceed in three interrelated directions. First, theoretical work will develop a model for movement generation on the basis of a network of coupled dynamic systems. Second, experiments with human subjects will study the specific features of movement trajectories in tasks involving multi-joint arm movements. Third, the model will be used to synthesize movements of a human-like robot arm in order to compare these model-driven movements with the movements of the human participants. In this way the model's validity and robustness will be tested in a real physical environment. In addition to increasing our basic understanding of complex movement generation the results should be applicable to the control of artificial devices such as robotic systems or limb prostheses, and may also contribute to the diagnosis and treatment of movement disorders.

This is the first year funding of a three-year continuing award. The goal of the "Virtual Trainer" is to create a computer-based animation that can interactively teach how to move. The hardware requirements for the Virtual Trainer in its final stage would be an inexpensive state-of-the-art personal computer equipped with a camera system. The Virtual Trainer will be able to demonstrate movements to its user, monitor the execution of these movements by the user, and suggest corrections in case of inadequate performance. The Virtual Trainer will be useful in a large number of applications, including rehabilitation of movement-impaired patients (e.g., stroke-patients), sport and exercise education, dance instruction, and interactive entertainment industry. Additionally, the technology developed for the Virtual Trainer has the potential to pioneer new algorithms for robot control using "teaching from demonstration", to contribute to the development of automated monitoring systems for human environments, to the generation of humanoid computer simulations, and also to gaining new insights into biological motor control and the functioning of the nervous system. The research team of this project will primarily focus on issues of movement recognition and movement generation with the Virtual Trainer for rehabilitating stroke-impaired patients with upper and lower limb disabilities.

ABSTRACT


Proposal #0312802
Title: ITR: Towards Organic Computing in Computer Vision and Robotics
PI: Schaal, Stefan
U. of Southern California


This project focuses on developing a methodology to advance a self-organizing computing paradigm in flexible robotics and vision in natural environments. These application domains require systems of tremendous complexity, as also indicated by the fact that a large part of our brains, with a processing capacity orders of magnitude larger than any technical system, is devoted to vision and control of behavior. It is doubtful whether systems of the required complexity can be designed by current computing methodologies. It is proceeded with the realization that radically new design principles will have to be developed, and are seeing this project as part of a broader effort to establish the new computing paradigm Organic Computing (OC). OC aspires to understand and emulate information processing in living systems, a form of information processing that does not seem to follow traditional algorithmic control, but rather mechanisms of evolution, adaptation, goal-oriented self-organization and learning.
The study will perform towards this goal by a series of theoretical and experimental studies that will demonstrate the principles of OC and how it can replace traditional programming in classical problems of computer vision and robotic control.
On the intellectual side, the project will contribute considerably to robust large-scale intelligent and autonomous systems in the future, and a better scientific understanding of the functional principles that are at the basis of autonomous vision, motor control, and visual-motor coordination, also with a view towards interdisciplinary exchange with the neuroscience. The broader impact will be to profoundly transform the style in which large software systems are developed and to open computer systems to the direct creative influence of the non-technical user. The significance of our particular sample applications will be progress with advanced sensing, perception and actuation systems in unstructured environments, as a basis for HCI systems and also for the emerging field of neuro-prosthetics and rehabilitation engineering as well as robotics and autonomous vehicle control.

Research in neuroscience and motor psychology has made tremendous progress in generating better understanding of how the human brain generates motor behaviors. At the same time robotics and computer graphics have created increasingly impressive examples of theories and implementations of a variety of movement behaviors, as seen in humanoid robotics and interactive animations and games. Despite this progress, however, a major shortcoming in all these disciplines remains an understanding of how complex movements that we make every day can be created and combined flexibly, robustly, and autonomously. The goal of this project is to develop and evaluate comprehensive information processing models of human motor behavior, to overcome these shortcomings.
The PIs will investigate the algorithms and representations (such as what is stored in long term memory) that enable the skilled behavior we see every day, using brain imaging, studies of behavior, and evaluations of our ideas on humanoid robots and in simulation. It brings together a set of researchers who have individually or in small collaborations addressed fragments of this challenge. The PIs have been successful in investigating individual and highly specialized motor tasks, but have not yet integrated a significant number of behaviors such that a robot or simulation could autonomously and robustly interact with a dynamic environment. Members of this team have built biologically inspired locomoting and
humanoid robots that balance; walk and run on both flat terrain, inclines, and stairs at a wide range of speeds; accurately place their feet while walking and running; jump and leap; jump through hoops; perform flips; recover from slips, trips, and stumbles; compliantly interact with humans; throw, catch, hit, and juggle balls; devilstick; and play air hockey. They have received equipment funding to develop a next generation humanoid in collaboration with Sarcos, from the NSF CISE Collaborative Research Resources (Research Infrastructure) Program. This humanoid experimental testbed will allow them to develop and evaluate their proposals as to how behavior is generated much more effectively.
In the past, this group and others have focused on modeling single tasks. This project focuses on developing and testing approaches to coordinate many behaviors, and handle behavior selection, multiple tasks, behavioral transitions, and error compensation, making the crucial step from highly specialized investigations to a more general theory of information processing in human motor control.

This project, acquiring a mobile humanoid robotics platform as the centerpiece of a Human-Centered Robotics Lab, aims at assisting a broad population in need, based on the belief that the most suitable form of multi-purpose assistive machine for humans will be human-like. This new kind of robot, not highly accurate, stationary, single task machine with sensing abilities as for typical industrial applications, is richly equipped with multi-model sensing, a high level of dexterity, compliance for safe operation, and mobility. Endowed with the appearance and behavior of a social system appropriate for human environments, it can perform a large number of assistive tasks, autonomously or in collaborative instruction with humans. A humanoid robot instigates a variety of original research. Developing humanoid behavior advances robotics and automation technology while promoting interdisciplinary interaction with natural sciences

This is funding to support a PI Workshop on the campus of the University of Southern California in Los Angeles, on September 27-28, 2006. HRI is one of the two cross-cutting technical areas defined in IIS Division's new solicitation entitled "Information and Intelligent Systems: Advancing Human-Centered Computing, Information Integration and Informatics, and Robust Intelligence" (NSF 06-572). The 1.5-day workshop will bring together current NSF PIs with ongoing HRI research programs, along with selected additional members of the research community with related interests, to help NSF identify emerging trends in this rapidly evolving field. Participants will discuss and prioritize the important subfields of HRI that the scientific community believes will have a major impact in the near-to-medium term. This workshop is one of several similar events, each focusing on a particular aspect of the new solicitation, being sponsored by IIS Division with the goal of encouraging the research community to provide input to NSF in its strategic planning process as we prepare for the challenges and opportunities anticipated during the coming period of explosive technological growth and change.

Broader Impacts: This workshop will present a unique opportunity for NSF PIs and other members of the rapidly growing human-robot interaction (HRI) research community to discuss the state of the art and to engage in strategic planning toward shaping the direction of this new and interdisciplinary research area with major implications on the future of robotics in society. The workshop organizers will produce a comprehensive workshop report, and the workshop website will become a permanent repository of the workshop discussions, breakout group reports, etc for general access by the research community at large.

The goal of "Skill Acquisition Through Interactive Avatars" is to create a computer-based humanoid simulation that can teach humans how to move. The hardware requirements for such as system in its final stage would be an inexpensive state-of-the-art personal computer equipped with a camera system. The avatar will be able to demonstrate movements to its user, monitor the execution of these movements by the user, and suggest corrections in case of inadequate performance. In order to be effective, the avatar will take into account neuroscientific knowledge about the organization of human motor control and human motivation during learning. This research will be useful in a large number of applications, including rehabilitation of movement-impaired patients (e.g., stroke-patients), sport and exercise education, dance instruction, childcare and special needs education, and interactive entertainment industry. Additionally, the technology developed for this project has the potential to pioneer new algorithms for autonomous robot control using "teaching from demonstration", to contribute to the development of automated surveillance systems for human environments, to the generation of humanoid computer simulations, and also to gaining new insights into biological motor control and the functioning of the nervous system.
As intellectual merits, it will be necessary to gain new understanding of how to recognize and classify human movement from real-time motion capture, how to create skilled movement, and how to teach humans effectively. The basic research problems of understanding human movement, both in terms of movement perception as well as movement generation, are central to advancing information technology in human-computer as well as human-robot interaction, i.e., the creation of autonomous artificial perception and movement systems. Research on intrinsic motivation in motor learning will advance important insights into how to help humans with learning disabilities, but also how to create machines that are motivating to interact with.
As broader impact, the research of this project will make important steps towards creating interactive environments that can assist people in their professional and private lives. Such technology will soon become a major component of our world, starting with clinical, entertainment, and business scenarios, and finally finding its way into private households. Understanding how to create and teach skilled movement in a user friendly way will be useful in building autonomous robot systems that can assist humans, entertain humans, replace humans in hazardous environments, rescue humans, or simply become a companion

This research addresses the question of how complex future robotic systems, e.g., like humanoid assistive robots, can acquire, refine, and maintain a variety of motor skills that enable them to operate autonomously in normal human environments. Humans excel in their abilities to perform motor skills due to various aspects, including i) imitation learning, which allows them to transfer prior knowledge about a task from a teacher to a student, ii) trial-and-error learning, which provides them with means to refine skills, iii) reactive behaviors, which can deal with dynamic and stochastic environments, and iv) compliant control, which is a basic mechanism for robustness against disturbances and promotes safety to act amongst other humans. Understanding the basic mechanisms of these abilities will lead to technological advances towards truly autonomous robotic systems.
Our technical work includes research on modular representations of motor control in terms of movement primitives, research on trial-and-error improvement of motor primitives and sequences of motor primitives with trajectory-based reinforcement learning using novel techniques from probabilistic reinforcement learning and path-integral reinforcement learning, research on reactive behavior using direct coupling of motor primitives to perceptual variables, and compliant control with the help of operational space controllers that can be learned. Besides traditional benchmark simulation studies, our evaluations will include the learning of motor skills with a full-body humanoid robot, a system that significantly challenges the scalability of our methods.

0922784
Loeb

This proposed MRI instrument will provide a highly flexible testbed for diverse research activities in the general field of robotic and prosthetic haptic. Robots have had a huge economic impact on certain types of industrial productivity in repetitive and/or hazardous environments but they are currently unable to handle common objects or tools. Providing robotic manipulation with haptic capabilities similar to humans would greatly extend the range of applications and environments in which they could be used.
There are three major aspects of the development project: i) Design and production of TAC modules; ii) Procurement of and interface with commercial mechatronic hand plus arm; iii) Development of software modules for compliant exploratory behaviors, processing of sensory data and extraction of information about external objects.

The proposed instrument represents a fusion of recent advances in biomimetic principles of sensory transduction, haptic exploration and compliant control.

In a not too distant future, assistive robots will become a natural part of the human society, in hospitals, schools, elder care facilities, inner city urban areas, and eventually homes. While wheeled robots, e.g., a humanoid torso on a mobile platform, can cover a range of tasks that assistive robots will be needed for, eventually, legged robots will be the most suitable, as legs increase the effective workspace of a robot and allow maneuvering more complex terrains like steps, curbs, and cluttered and rough terrains in general.
This project investigates biped locomotion with a Sarcos humanoid robot. In contrast to most other projects in biped locomotion, it emphasizes walking over uneven and rough terrain, obstacle avoidance, recovery from unexpected perturbation, and learning methods for motor control, as these issues seem to be the most important for working in dynamic and partially unpredictable human environments. Another focus is on dexterous movement control, i.e., control with a maximal amount of compliance and minimal negative feedback gains, using advanced operational space controllers with internal model control. Dexterous, compliant control will increase safety of the robot when accidentally impacting with humans or obstacles, and it will also allow the robot to recover more easily from external perturbation simply by ?giving in?. Such a control approach requires departing from the traditional high-gain position controlled humanoid systems, and focuses on torque control, reactive instantaneous control instead of finite horizon optimization, as well as efficient motion planning and learning methods.

"This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5)."
Proposal #: 09-60061
PI(s): Kim, Youngmoo E., Gogotsi, Yury, Hong, Dennis H., Regli, William C., Schaal, Stefan
Institution: Drexel University
Title: Development of Common Platform for Unifying Humanoids Research
Project Proposed:
This project, developing and disseminating HUBO+, a new common humanoid research platform instrument, enables novel and previously infeasible capabilities for future research efforts while working with a common instrument. HUBO will be the first homogeneous, full-sized humanoid to be used as a common research and education platform. Eight universities (Drexel, CMU, MIT, Ohio State, Penn, Purdue, Southern California, and VaTech), representing a critical mass of humanoids research within US, participate in this development of the world's first homogeneous full-sized humanoid team. Building upon unique expertise, the work extends current capabilities, resulting in six identical units, facilitating the following potentially transformative advances in robotics:
- A state-of-the-art, standardized humanoid platform instrument with embedded capabilities for sensing, manipulation, and rapid locomotion, ideal for a broad range of future humanoids research
- The ability, for the first time, to directly compare and across validate algorithms and methodologies and consistently benchmark results across research teams
- Novel energy storage technology for mobile robotics incorporating supercapacitors for operations requiring high power density, far exceeding the capabilities of traditional battery-only power sources
- A widely distributed platform that motivates, recruits, and trains a broad range of students spanning multiple disciplines, including artificial intelligence, digital, signal processing, mechanics, and control
Humanoids, robots engineered to mimic human form and motion, open broad avenues of cross disciplinary research spanning multiple fields, such as mechanical control, artificial intelligence, and power systems. Common humanoids are rarely autonomous and are not-ready for unconstrained interaction with humans. The most compelling demonstrations are meticulously pre-programmed and painstakingly choreographed. A few common platforms have already advanced some research. Hence, having a consistent platform should facilitate rapid progress in areas needed for autonomy and natural interaction, including mobility, manipulation, robot vision, speech communication, and cognition and learning. However, although currently Japan and Korea are considered world leaders in design and construction of humanoids, best practices have not been developed for constructing multiple, identical humanoids. These conditions call for the making of an urgently needed benchmark providing evaluations and cross-validation of results. With this development and the servicing of 6 humanoids, this project aims to create knowledge and best practices contributing to robotics research, possibly leading to the standardization needed for ubiquity.
Broader Impacts:
The instrument enables US researchers to develop expertise in the design and construction of humanoids, while the distribution of the work activities ensures the broad dissemination of the knowledge. Humanoids research, inherently interdisciplinary and integrative, inspires young students. The graduate and undergraduates students participating are likely to receive a world-class training in robotics. Outreach partners, including several high-profile museums will introduce people of all ages to the exciting technologies of robotics, particularly useful in recruiting K-12 students into science, engineering, mathematics, etc. A partnership with the Science Leadership Academy (SLA), a magnet school with more than 63% underrepresented students, assures their involvement. With SLA, the project initiates an annual program modeled on a NASA-style experiment design competition, in which students use simulation tools to propose humanoids projects and activities. Selected winner(s) will have their proposed projects implemented on HUBO.

This project focuses on tackling a critical barrier to long-term autonomy for robotic systems, namely the lack of theoretically well-founded self-calibration methods for inertial and vision-based sensors, commonly found on sophisticated robots. The project is motivated by the vision of power-up-and-go robotic systems that are able to operate autonomously for long periods without requiring tedious manual sensor calibration. The research team addresses this problem in the context of vision-based mobile manipulation and navigation. The core foci of the work are: 1. the development of a unified mathematical theory of anytime, automatic calibration for visual-inertial systems, and 2. an experimental characterization of the resulting algorithms with state-of-the-art, sophisticated robots of significant diversity (humanoids performing mobile manipulation and autonomous ground vehicles navigating outdoors). Inertial sensing is critically important for humanoid balance control, while visual sensing relates the 3D world to the robot's body coordinates thereby enabling manipulation. In the case of autonomous ground vehicles, monocular and stereo camera calibration is still commonly performed manually using a known calibration target. The project obviates the need for this requirement. The expected outcomes of the project are: 1. a theoretical foundation for humanoid robots to function autonomously in unstructured environments over significant periods of time, and 2. new navigation algorithms for ground vehicles allowing them to see further with greater acuity. The project explicitly incorporates undergraduate research in cooperation with an REU site currently operational at the USC Computer Science Department.

Cognitive robotics has been identified as a research area of common interest to Japan and the US; in addition, it is of mutual benefit to both countries to explore areas for collaborative work where each side brings complementary strength to the project. Moreover, there is already a good start in collaborative research among some US-Japan roboticists. We propose to propel this joint cooperative research to a higher level of impact by holding a joint NSF-JST workshop which brings together the best researchers from both countries and have them discuss and exchange ideas about the most opportune research directions, then work at a deeper level on some specific challenges in these areas in order to be well-positioned for a next round of collaborative funding opportunities.

This workshop has the potential to result not only in new research directions which could ultimately impact many industries (e.g., elderly care, logistics, security, medical, exploration, monitoring, etc.), but also lead to broader international opportunities for faculty and students. The goal is to have the results of this workshop result in high-impact collaborative US-Japan research with student and faculty exchanges and long-term visits. This renders our students and faculty more globally aware and competitive.