Humanoids 2018 Workshop
Humanoid Robotics: Dead-end or Bright Future?
Humanoid robotics has a great potential to solve various important problems that the modern society faces, such as rapid growth of human population, percentual increase of older people, pending natural and human-made disasters, unsafe working environments for humans, etc. While the research community has made excellent research, engineering and advancement in humanoid robotics, the optimistic past predictions and expectations of humanoid robots helping us in various aspects of our daily lives have not yet been fully met and some still seem far. Recently, events such as DARPA Robotics Challenge gave a rather harsh demonstration and realization to both the research community and general public about the actual state of the current progress in humanoid robotics with respect to predictions and expectations. The harsh realization that came from events like DARAP Robotics Challenge may be attributed to overly optimistic past predictions, and to some degree unrealistic and not-critical-enough discussion among the research community.
Nevertheless, there are still many bright and positive points that we can cling to and should focus on. For example, recent advancements such as those demonstrated by Boston Dynamics gave us some great boost. Even though the optimistic goals with respect to applicability of humanoid robots have not yet been fully realized, the humanoid robots and related research is still being used to many ends. For example, the humanoid robotics research can contribute to understanding of humans and their behavior. Furthermore, there have already been some commercial applications in social human-robot interaction (e.g., Pepper, etc.), ranging from education to marketing. Last but not least, emergence of hobbyists in humanoid robots is also promising.
The aim of the proposed workshop is to have a constructive and above all realistic discussion regarding the progress, expectations and applicability of humanoid robotics, and to examine both positive and negative aspects alike. First we want to discuss the nature of overly optimistic past prediction and expectations. Then we want to re-examine the excellent research that led to the existing state-of-the-art and find the most effective ways get to the expected aims. We will identify the main bottlenecks in the progress and try to find solutions or/and workarounds. Finally, we will critically re-evaluate the possible applications of humanoids and discuss their strengths/weaknesses with respect to competitive (classical) solutions.
Some of the questions we want to address and discuss
How do you see the current progress of humanoid robotics with respect to optimistic predictions in the past?
- How is AI affecting the research in humanoid robotics?
- (How) does human neuroscience benefit from the research in humanoid robotics?
- What and where are the main bottlenecks that prevent (expected) progress: hardware vs software?
- How to make humanoid robots feasible for customers in respective markets to be interested in humanoids compared to alternative (classical) solutions?
- Household tasks: humanoid robot versus specialised robots/devices (sweeping robot, dishwasher, etc.).
- Disaster response: humanoid robot vs quadrupeds/hexapeds/wheels/etc.
- Military application: has military given up on humanoid robots? Do we even want military humanoids?
- Entertainment industry: potential savior of humanoid robots?
- Development of hobbyists: how we can collaborate with growing number of hobbyists?
- What can we do or change to speed up the progress towards achieving applicable humanoid robots?
Conclusions from the workshop
- While there has been a significant progress in humanoid robotics, it may take a while to have applicable humanoid robots in our daily lives.
- Many videos/papers of research and development in humanoid robotics are misleading to the general public and produce illusion that the current state-of-the-art is much higher than it actual is. For example, if the robot needs many unsuccessful attempts before there is one successful attempt, which is used in the released video/paper, the public (and even research community) may get an impression that what is in the video is the actual state-of-the-art. However, there is still a long way to achieve the same result with a decently high success rate. Since the public and the funding agencies may be misled to think that the problem is already solved, they might not be willing to further fund the crucial progress towards achieving high success rates. Therefore, the success rates should be clearly reported in the released videos/papers.
- It took 30 years to develop applicable, reliable and commercially available torque-controlled robotic arms. Humanoid robot is much more complex and includes many elements, therefore it may take some more time.
- However, if it takes 100 years it might be pointless. When it comes to application of humanoids, we should first make smaller steps (i.e., find a niche, and very specific and limited applications / business models) in order to be able to achieve at least some initial commercial success. If not, the society and the founding agencies might give up on humanoid robots and stop supporting the research in the long-term.
- The aircraft manufacturing industry might be a good example of application of near-future humanoid robots. This kind of industry requires complex operations and is relatively rich enough to be able to afford such robots. Another good example is entertainment industry and application of humanoid robots as toys.
- Quite a lot of development has been done on hardware and control. However, perhaps we do not use it enough or properly. For example, not much has been done in fall reaction and recovery. Even with the best hardware and control falls will happen and therefore the ability of reaction and recovery is fundamental.
- Hardware should be robust and resilient. While classical electric-motor based actuators and control might still be useful in future, the progress should be made towards variable impedance actuators and control.
- Artificial intelligence is still far behind. A high-level controller with exceptional cognitive capabilities might be able to use the existing hardware and control to make various complex tasks that cannot be done at the moment. For example, if a human is included into the robot’s control loop through a teleoperation or prosthesis, it can perform well with the existing technology, due to the superior cognitive capabilities (i.e., ingenuity, adaptability, problem understanding, prediction, etc.).
- Hobbyists and amateur researchers can potentially contribute to the progress, and could provide encouragement (or even a positive competition) to the professional researcher. On the other hand, they could potentially make professional researchers look bad if they can achieve great results with low funding. Nevertheless, we should also pay attention to any fake videos.
- We need more healthy competition between companies that are producing humanoid robots.
- For disaster response scenarios, a combination of wheels and legs might be the best solution.
The workshop will be held in Meeting Room 3
|09.00 - 09.30||Introduction by the organizers (Luka Peternel)|
|09.30 - 10.00||Talk by Tamim Asfour|
|10.00 - 10.30||Coffee Break|
|10.30 - 11.00||Talk by Giorgio Metta|
|11.00 - 11.30||Talk by Francesco Ferro|
|11.30 - 12.00||Talk by Nikos Tsagarakis|
|12.00 - 12.30||Talk by Christian Ott|
|12.30 - 14.00||Lunch Break|
|14.00 - 14.30||Talk by Pierre Gergondet (on behalf of Abderrahmane Kheddar)|
|14.30 - 15.00||Talk by Shuuji Kajita|
|15.00 - 15.30||Talk by Sven Behnke|
|15.30 - 16.00||Coffee Break|
|16.00 - 18.00||Roundtable Discussion|
Italian Institute of Technology, Italy
Jožef Stefan Institute, Slovenia
Karlsruhe Institute of Technology, Germany
Title: Humanoid Robotics: Key Science and Technology for Understanding Intelligence
Abstract: The major motivation and the common problem tackled in humanoid robotics research is understanding how to implement human-like intelligence in technical systems. Such understanding relies on human-like embodiments with rich perceptual, motion and cognitive abilities which enable physical interaction with the real world and learning from interactions. The talk will discuss current progress in the field and the role of humanoid robotics as science and technology of creating human-like robot systems with cognitive abilities and human-like motion intelligence leading to better understanding of intelligence.
Title: Humanoid Robots: From Fiction to Reality?
Abstract: Humanoid robots have been fascinating people ever since the invention of robots. They are the prototypical embodiment of artificial intelligence. While in science fiction, human-like robots act autonomously in complex human-populated environments, in reality, the capabilities of humanoid robots are still quite limited. In the talk, I will review the history of humanoid robots from a personal perspective, discuss the state-of-the-art, and speculate about future developments in the field. In particular, I will report on some of the developments of our team NimbRo, which competed with great success in humanoid robot competitions, such as RoboCup Humanoid soccer, RoboCup@Home, DARPA Robotics Challenge, and DLR SpaceBot Cup. Our team also used humanoid robots for intuitive multimodal communication with humans and for complex locomotion and manipulation tasks in disaster-response scenarios. Even though much progress has been made, current humanoid robots lack sensing, perception, planning, learning, and actuation capabilities to come even close to human performance. I speculate that affordable and capable humanoid standard platforms would be helpful to overcome some of these limitations.
Title: Progress achieved in humanoid robotics and future challenges
Abstract: PAL Robotics' humanoid robots are currently used in the service and entertainment industries, as well as for research in the fields of collaborative robotics for Industry, Assisted Living and also Smart Cities. This talk will give insights into the achievements and challenges ahead of humanoid robots based on such diverse experiences. The talk will also explain how PAL Robotics has been designing and building humanoid robots, and the challenges and bottlenecks that the team overcame from 2004, when the team started, up until the development of TALOS, the latest biped robot from the company. A special highlight will be put on TALOS since the torque-controlled robot integrates groundbreaking technology that opens a new scope of applications for humanoid robots.
Title: Hobbyists and professional research community
Abstract: Thanks to the advance of cheap computers and other devices, it is becoming possible for ordinary people to build humanoid robots at affordable costs. I will introduce some impressive developments done by hobbyists. Then, their positive and negative effects are discussed. In my opinion, we should never neglect their activities. We can expect a blight future of robotics by establishing bi-directional communication between hobbyists and professionals.
Title: Humanoid Robots for Aircraft Manufacturing @AIRBUS
Title: Possible business models for humanoid robots
Abstract: Humanoids pose incredible challenges because of their complexity. Walking on two legs is tricky, manipulation is still limited, and if we’re strict about sensing modalities, then human-like sensors are not necessarily the easiest to deal with (think of the difference between plain stereo vision and RGBd cameras). However, humanoids have certain advantages in environment designed for human beings, for example, using tools designed for humans, exploiting the environment’s affordances, interacting naturally with people. These are all active areas of research. I will speak about some of our research results in these different areas and provide some of the take home messages we distilled from working for more than 10 years on the iCub humanoid project. My claim is that a bright future is possible, assuming we keep ambitions proportionate to the technical difficulty and we are well focused on identifying the “business model” (so to speak) for humanoid robot applications to thrive.
Title: Torque controlled humanoids
Abstract: During the last ten years several new humanoid robots have been presented, which allow joint torque control as an alternative to classical position control. The torque interface in these robots is used for implementing force based control approaches, like whole body impedance and compliance control. In this talk I will first give a short overview of these developments. Then I will present some recent results made with the robot TORO at DLR in the context of the EU project COMANOID in which humanoid robots are evaluated in an aircraft manufacturing task. Finally, I will give an outlook on the next generation of physically compliant humanoids.
Title: Humanoid Robotics : Challenges and Work Progress @ IIT-HHCM