I attended ERF2017 in Edinburgh last week 22.-24.3.2017.
My interest is applying robotics for healthy aging and assisted autonomous living, so I followed tracks on Artificial Intelligence (AICoR), Health, Social Impact, Ethical Legal and Societal.
I did not follow Industrial, Agri-Food, Harsh Environments, Space, Underwater, Maintenance and Inspection, Logistics, Construction, Neurobotics, Miniaturized, System Engineering, Innovation, Awards or Competitions tracks, except for what was presented in the opening session and Thursday morning success stories session. See the program at www.erf2017.eu.
The official opening included introduction talks from Eu Robotics leaders Dr. Reinhard Lafrenz, Dr. Bernd Liepert, the Edinburgh Center for Robotics hosts Prof. David Lane, Prof. Sethu Vijakumar, and Dr. Frank Broz, EC department of AI and Robotics head Dr. Juha Heikkilä.
Europe is leading continent in Service Robotics
OPENING KEYNOTE – Member of Scottish Parliament Keith Brown gave the view on robotics from Scottish government and Cabinet Secretary of Economy Jobs and Fair Work point of view. Key introductory message is that Europe is leading continent in Service Robotics in terms of the 38% global share and number of startups. EU has the biggest research program in robotics, AI is more relevant than ever and AI is part of all projects funded, EC is proceeding with robotics Ethical and Legal rules with the EU Parliament decision in this February (20170210IPR61808), and participation in EFR is steadily increasing – 750 participants this year. Next conference in in Tampere Finland 2018!
KEYNOTE 1 – Mr. Tim Ensor CCO of Tharsus, the leading UK autonomous system robotics designer and manufacturer, delivered a message to strive for simplicity is design to achieve the system reliability, and fast prototyping to meet customer requirements. As an example he showed Ocado Smart Platform (http://www.tharsus.co.uk/case-study/ocado-smart-platform-robot/) robotic handling system for automatic pickup of retail customer groceries in minutes.
KEYNOTE 2 – Dr. Raia Hadsell, Senior Research Scientist at Google DeepMind, delivered a strong message for deep learning and reinforcement learning in robotics. She argued that following deep network learning breakthroughs in Speech Recognition 2010, Computer Vision 2012, and Machine translation 2014 there would be a similar development in Robotics very soon. The biggest challenge is that deep reinforcement learning is still very data inefficient and hence slow. It can be sped up by transferring the training in simulation results to real robots – which is hard to do in practice. An alternative method is to use progressive neural networks to transfer learning from similar tasks to new tasks, which can be done in minutes versus hours. See http://www.raiahadsell.com/ for more details.
KEYNOTE 3 Mr. Stan Boland, Five AI CEO, gave an inspiring keynote on developing superhuman perception and prediction AI to deliver safe urban autonomous personal mobility. Europe has the talent and the funding for what takes to the leader in this space. Five AI is well on its way to piloting their platform with the City of London. See http://www.five.ai/news/ for interesting stuff.
AI and COGNITION in ROBOTICS (AICoR): The first AICoR workshop was before the official opening on Wednesday. Session was opened by Alessandro Saffiotti from Örebro University after which Cécile Huet the deputy head EC unit on Robotics and AI set the scene for the workshop: European efforts are scattered, there is huge global investment in robotics, AI is not exploited enough → join forces, focus on strengths, deliver tools for AI utilization for the benefit of EU citizens in healthy aging – among other things. Challenges learned from project ROMEO and Shadow Robotics were presented next. Users want social robots to be more intelligent, understanding and relevant, context and situation aware, more reliable, predictable, trustworthy and allow the user to stay in control via easy to use human robot interface (HRI). From development point of view the control programming should move to teaching and learning by higher level instructions, cognitive capabilities and knowledge sharing. Finally the workshop discussed how cognitive actions would be learned and defined, and how much freedom/autonomy the robot should have deviating from the script and selecting its goals.
The second session for AICoR presented a number of projects where AI and cognitive issues have been addressed. In robotics the intelligence/control is talked about as planning being done by the planner. The planner needs to be in closed loop with the execution, but in addition it needs to take humans in account and be able to repair/modify the plan as needed. Robots need to ask human for advise (symbiotic autonomy, dynamic dialogue), prioritize competing plans according to the mission perception, and share knowledge and skills with other robots. Horizon 2020 CARESSES (http://www.caressesrobot.org/) objective is to build culturally competent care robots, able to autonomously re-configure their way of acting and speaking, when offering a service, to match the culture, customs and etiquette of the person they are assisting. Project is easy to confuse with CARESS (http://www.project-caress.eu/home/) which is an ongoing Erasmus+ funded project with Omnia, SuPer and Ministry of Education participating is promising guidance for robotics needs on the healthy aging and home care applications. The SQUIRREL (http://www.squirrel-project.eu/home.html) project is also interesting for home care as it is solving challenges of dealing with clutter (mess) in unstructured domestic setting.
The second workshop on AICoR focused on how to best apply robotics with AI and cognitive capabilities. The questions addressed in this workshop were 1) The demand – what cognitive capabilities make a difference, 2) The offer – what AI methods and tools are mature enough to apply, 3) The gap – what further development is needed, 4) The infrastructure – how to match the demand and offer, and finally 5) Concrete examples on applications of AICoR . See http://aass.oru.se/Agora/ERF2017/index.html for workshop details. The expert statements before the group discussions identified a lot of challenges. For robots in healthy aging at home the normal environment differs from industrial harsh yet controlled environments. Human cooperation is now a common challenge and required perceiving and understanding human behaviour together with robotic planning and action. Flexibility and adaptability of robots is an issue, learning is slow, and programming for specific applications is an obstacle due to scarcity of skilled programmers. Automatic abstraction of the cognitive inputs and goal based reasoning for action is a definite need in care applications as well. The RoboHow project (http://robohow.eu/project) is an example of what is needed: a cognitive robot that autonomously performs complex everyday manipulation tasks and extends its repertoire of such by acquiring new skills using web-enabled and experience-based learning as well as by observing humans. See http://actioncores.org/ and http://www.open-ease.org/ for tools in modelling planning manipulation tasks. For the home care robot there is still work to do for personilised interaction where user specific activities goals and psychological state can be recognized, and robot can reason and how it is capable of cooperating with the user.
Friday morning session on AICoR focused on combining IoT with AI and Robotics. The expert statements addressed the IoT added value in robotics solutions, what AI and cognitive functionality has been used, what technology is considered mature enough for application needs, what are the obstacles for the integration, and finally any learnings to share. The added value of IoT is through data acquisition from low cost sensors and enabling the distributed sensor fusion for AI analysis, and for sharing the results for collective intelligent actions. Proactive maintenance can be made more efficient and the need for reactive measures is reduced – this applies to humans as well as machines. Integration is challenging due to the variety of sensors and low local computing capacity versus the cloud capacity. Robotic action relies on low latency planning which is a challenge to general purpose cloud platform services. The ACCRA Agile Co-creation of Robots for Ageing project (http://cordis.europa.eu/project/rcn/207079_en.html) was presenting their IoT integration thoughts. They’ll be applying FIWARE (https://www.fiware.org/) platform integrating universAAL (http://universaal.sintef9013.com/index.php/en/) features, Buddy and Astro robots. Crossdomain – IoT-AI-Robotics – integration is an architectural challenge to be solved.
ROBOTICS IN HEALTH – I attended three of the eight Robotics in Health sessions. The most interesting for healthy aging and home care was the session on European service robotics for eldercare and assisted living on Thursday afternoon. The HOBBIT (http://hobbit.acin.tuwien.ac.at/) was an FP7 project that has followup in H2020 by RAMCIP (http://www.ramcip-project.eu/ramcip/), RADIO (http://radio-project.eu/) and Enrichme (http://www.enrichme.eu/wordpress/) projects. The arm for picking up things was the most appreciated feature of the HOBBIT, but arm could not go under furniture which would have been desirable. Also speech control would have been appreciated in addition to touchscreen and elementary gesture control. The large size of the robot was challenging in some home environments. The system cost is below 16kEUR.
RAMCIP is aiming to adopt existing technologies from the robotics community, fuse those with user-centered design activities and practical validation, trying to create a step-change in robotics for assisted living. The RAMCIP vision is worth checking http://www.ramcip-project.eu/ramcip/content/ramcip-vision – discreet personal assistance and coaching to help maintain and even improve health of the person with impairments living home.
RADIO is using an integrated smart home/assistant robot system as the sensing equipment for health monitoring. The robot is able to unobtrusively become part of their daily routine, collect observations regarding activities of daily life (ADL) and mood, and use these observations to establish patterns and identify deviations. Multimodal analysis techniques will be developed for detecting Activities of Daily Life, taking care that they are directly applicable to common robot sensors. The robot will also recognize the mood of the user by means of both speech and facial characteristics analysis.
ENRICHME tackles the progressive decline of cognitive capacity in the ageing population proposing an integrated platform for Ambient Assisted Living (AAL) with a mobile service robot for long-term human monitoring and interaction, which helps the elderly to remain independent and active for longer. The ENRICHME system is an interactive mobile robot in an assisted living environment for the provision of advanced user services, integrated within a domestic RFID ecosystem. It has three different levels of intelligence: Robot Intelligence, Ambient Intelligence, and Social Intelligence. Check http://www.enrichme.eu/wordpress/about/our-approach/ for details.
i-support (http://www.i-support-project.eu/) project addresses the ”Bathing/Showering (d510)” ICF core set in a manner that respects dignity, self-esteem, control and reduces cost and need for supervision. It targets to develop a service robotic system for bathing tasks.
In the workshop discussion the costs and the access to the assistive solutions revealed significant differencies between EU national policies and standards of living, and also a some disconnect between the robotics research and public health services reality. I think adaptation of new technology will be by wealthy individuals first through private care service providers and public services will adopt new technologies slower.
Friday morning health session was on Ethical Legal and Societal Aspects (ELSA) of Healthcare Robotics. Themes of liability, equality, robot autonomy and personhood, de-humanisation, deception, manipulation, and privacy of personal care robots was discussed. The MOVECARE (http://cordis.europa.eu/project/rcn/206414_en.html) project ELSA impact was presented as introduction to further discuss service robot critical issues. Harm and benefit is one way to solve ethical questions, yet the impact of service robots is not well understood. On the high level the guidance is clear to promote autonomy, dignity, social participation, and safety in healthy aging context.
There was three sessions on Success Stories of which I attended the one on step change results on Thursday morning. Two of the eight presented completed projects were interesting for home care robotics. The STRANDS project (http://strands.acin.tuwien.ac.at/index.html) key accomplishment was the robustness for long term autonomous operation in everyday environments and the learning from experience with human robot interaction in a care facility. The other relevant project is Xperience (http://xperience.org/) where robots are bootstrapped through learning from experience using a structural bootstrapping idea taken from child language acquisition research. The affordance reasoning results of this project are applied in the SQUIRREL mentioned in AICoR part.
SOCIAL ASPECTS – There was four sessions on Social impact of robotics of which I attended the one on Thursday morning. This session was on barriers to market for social robots in contemporary applications. Social robots are used in humanly structure environments, so algorithms and behaviour developed in industrial structure or harsh unstructure environments do not apply as such.
The cultural and human behavioral requirements need to be considered, and assistive robots get easily humanized. The humanization of robots is evidenced by sosiological and psychological studies on emotions for social robots. Social robot behaviour is seen as childlike and cute when clumsy, scary when unexpexted movements are happening, sad and pityfull when robot is inactive or down. People want to remain in control and maintain autonomy in carrying out certain tasks, so robots are best seen in assistant roles. If the robot has no understanding of the social context, its behaviour can be unacceptable. People exposed to robots need to be observed to learn about HRI requirements, and attitudes towards robots. It is utmost important for acceptance that the robot immediately provides concrete value for the users instead of just being a toy creating a diversion from carrying out pressing tasks. Robot should be autonomous enough, and adaptive, so that is does not require constant support.
Risto Soila
Risto is RAG’s Innovation Guy, R&D professional focusing on innovation for healthy living by applying IoT, Digitalization and Robotics ideas. You can contact Risto by email: risto.soila (a) rag.fi