In the dynamic field of robotics and artificial intelligence, we’re excited to share a conversation with Stephane Doncieux, Professor of Computer Science at the Institute of Intelligent Systems and Robotics (ISIR) at Sorbonne University. As the Principal Investigator for the PILLAR-Robots project at Sorbonne, Professor Doncieux is at the cutting edge of developing robots that can seamlessly interact and learn within human-centric environments. Today, we delve into his perspectives on the evolution of robot autonomy and the collaborative potential within the field of Mobile Manipulation.
What does Pillar-Robots envision for the future of autonomous robots, and how do you see this technology evolving, also in the sphere of Mobile Manipulation?
Today’s robots can hardly deal with the variability of our environments if they must manipulate objects, no matter how simple the objects and the manipulations are. Something as simple as grasping an object, which seems trivial to us, is still out of reach beyond specific grippers that make the task trivial or beyond controlled conditions. Developing mobile robots with even simple manipulation skills, but that work robustly in everyday environments without the need for an expert to precisely define what needs to be done in every environment, would open many possibilities in service or industrial applications. Pillar-Robots aims to go in that direction by providing robots with adaptation capabilities based on open-ended learning, i.e., learning beyond initially known conditions.
Having robots that can adapt to new tasks with minimal expert involvement would be a great progress to make robots useful, but it raises the challenge of aligning this adaptation capability with what human users want: Ensuring the robot learns what the human wants it to learn poses a significant challenge. Current learning algorithms do not address this issue as learning is preceded by a preparation step in which an expert defines what the robot knows, what it can do, and how both aspects can be connected. This step allows the expert to define what the robot can learn. If we go beyond that and let the robot identify these different pieces of knowledge on its own, then we need to find a way to include users’ constraints and requests. As the goal is also to reduce the level of expertise required to operate the robot, we need to find intuitive interaction means. In Pillar-Robots, we propose to use natural language or gestures to fill in this gap.
The future of autonomous robots and of autonomous mobile manipulation robots is thus to have robots moving around in our everyday environment. Human users will talk to them or show them what to do, and the robot will achieve it without the need for an expert to adapt the robot program to the new task.
Looking into the future, Sorbonne is tasked with the Industrial Application Use Case in PILLAR. How do you see the industrial scenario unfolding and how the robotics and scientific community can help make this a success?
The scenario is organized around two different and complementary tasks. The first one is basically a pick and place task in which humans can request what they want through language. The main focus in this first task is grasping. This is clearly one of the main bottlenecks right now. If we can develop algorithms that allow a robot with any kind of gripper to grasp any kind of object and move it to another place according to robot user requests, we will make a significant step towards versatile mobile manipulation robots. The second task involves other kinds of manipulations than grasping: button pushing, moving a slider, plugging a probe, and making a measurement autonomously. This task is representative of what a robot could do in a small factory or in a recycling facility, but the manipulation it implies would also be useful for robots that do the housework beyond just cleaning the floor.
An important question is then to define what would make the project successful. How can we evaluate PILLAR robots on these tasks? There are many different performance indicators. Performance-based indicators are interesting and probably necessary, but they are not sufficient. Other criteria measure cycle time, safety, or energy consumption, but they do not evaluate what PILLAR is focused on. The goal of PILLAR is to develop methods that make robots easier to use. The performance indicators thus need to measure it. We have proposed performance indicators based on the versatility of the robot, defined as the number of tasks that can be achieved without reprogramming the robot, on the time to be operational, that measures how long it takes to make the robot solve a new task and end-user investment. These indicators are not straightforward to measure. We will certainly propose updates to make them easier to evaluate. The robotics community can help in refining these indicators and using them to better compare new algorithms under the perspective of robots’ ease of use.
In the last European Robotics Forum in Rimini, you highlighted the possible links between PILLAR and the Network of Excellence euRobin. Could you tell us how the cooperation between these two projects may support the mobile manipulators research field?
euROBIN aims at making the European robotics community join effort to develop robots that reach unprecedented levels of performance and autonomy. Many European projects have proposed original and innovative solutions. euROBIN aims at making people collaborate to combine these solutions. To do so, euROBIN has promoted the concept of “coopetitions,” i.e., cooperative competitions. The idea is to participate in one of the proposed challenges: outdoor robotics (bringing a parcel to the door), service robotics (opening the parcel and putting the objects it contains in the cupboard, emptying a dishwasher, …) or industrial robotics (performing a sequence of simple but accurate object manipulations). Instead of a classical competition in which every competing team arrives with its own solution, participants are encouraged to reuse others’ modules and to provide their modules to others. The goal is not to have someone alone that solves everything, but to have people share their work and integrate others’ work. The modules developed within PILLAR can be proposed to others that will test them and include them in their own pipeline, thus ensuring that PILLAR’s results are used in the robotics community.
This cooperation model could greatly benefit the mobile manipulators research field as different teams of researchers will develop new modules, test and combine them. The outcome will be a set of modules ready for integration, facilitating and speeding up research work, but also the work of industrials interested in developing robots for these use cases.
Pillar is working with the IMOL community to establish links and synergies with robotics and AI stakeholders across Europe. What would you say to European industry players and research groups who would like to know more about the project’s goals? Do you see a path to strengthen cooperation and produce meaningful results in the years to come?
The IMOL community is the research community that focuses on the research questions considered in PILLAR. IMOL stands for Intrinsically Motivated Open-Ended Learning. You can join the community to know more about the related events, and participate in them. See the community website for more details here.
The development of robots with strong adaptation capabilities raises many challenges. It is important for us to include all stakeholders as soon as possible to see what to focus on first and to better see what capabilities are the real game changers. We have set up a survey to collect feedback on these issues. Everyone is invited to fill it in here! Your feedback is very important for us!