Robolog 2017 workshop

This talk addresses some key decisional issues that are necessary for a cognitive robot which shares space and tasks with a human. We adopt a constructive approach based on the identification and the effective implementation of individual and collaborative skills. The system is comprehensive since it aims at dealing with a complete set of abilities articulated so that the robot controller is effectively able to conduct in a flexible manner a human-robot collaborative problem solving and task achievement. These abilities include geometric reasoning and situation assessment based essentially on perspective-taking and affordances, management and exploitation of each agent (human and robot) knowledge in a separate cognitive model, human-aware task planning and interleaved execution of shared plans.

Short bio.
Dr. Rachid Alami is Senior Scientist at CNRS. He received an engineer diploma in computer science in 1978 from ENSEEIHT, a Ph.D in Robotics in 1983 from Institut National Polytechnique and an Habilitation HDR in 1996 from Paul Sabatier University He contributed and took important responsibilities in several national, European and international research and/or collaborative projects (EUREKA: FAMOS, AMR and I-ARES projects, ESPRIT: MARTHA, PROMotion, ECLA, IST: COMETS, IST FP6 projects COGNIRON, URUS, PHRIENDS, and FP7 projects CHRIS, SAPHARI, ARCAS, SPENCER France: ARA, VAP-RISP for planetary rovers, PROMIP, ANR projects). His main research contributions fall in the fields of Robot Decisional and Control Architectures, Task and motion planning, multi-robot cooperation, and human-robot interaction. Rachid Alami is currently the head of the Robotics and InteractionS group at LAAS.

During the course of long term cooperation with humans, robots can potentially accumulate extensive records of their experience. The question is whether and how this information can be transformed into useable knowledge that has meaning that can be executed and/or re-used. I will report on our recent experience in how knowledge from human robot interaction can be extracted into logical forms that can be used for planning and reasoning, and how additional human input can enrich such representations to include causal relations between mental states that were not visible in the initial experience.
Short CV: Dominey is a CNRS Research Director in Lyon, INSERM U1208, responsible for the Human and Robot Cognitive Systems Team. After initial training in computer science and cognitive systems in the College Scholar Program at Cornell University, he worked as a software engineer in distributed OS design at Data General in Westboro MA, and then as a systems engineer in the Flight Project Support Office at JPL in Pasadena CA. At that time he completed the MSc. and PhD in computer science at the University of Southern California with Michael Arbib, and then began post-graduate studies with Marc Jeannerod at INSERM in Lyon. His laboratory combines human neuroscience, neural network modeling and cognitive robotics. In several EU and ANR projects he has implemented cognitive systems that allow cooperative interaction between humans and robots, emphasizing the role of language and autobiographical memory.

Automated Planning (AP) is the branch of Artificial Intelligence pivoted on the generation of plans: a series of actions guiding an autonomous system from the current state to its goal, accomplishing a required task in the process. Each action requires a certain number of preconditions to be fulfilled in order to be applied to a particular world state. Upon application, each action changes the world state with its induced effects. These actions can be represented by symbolic models: the blueprint of the domain-specific actions. AP came into light when research in the fields of operations research, theorem proving and studies into human problem solving started gathering steam, with the intent of solving problems that were being posed by the field of robotics among many others.
Knowledge Engineering for AP is the process that deals with acquisition, formulation, validation and maintenance of action models. The main issue for encoding action models is that they require a specific expertise, which can be acquired in one way by the help of a domain expert. This expert can introduce some human-induced error in the encoding due to his limited knowledge of the real world domain (HRI for instance). In these situations, the success of the AP systems fully depends on the skills of the experts who define the action model.
While it is possible to codify action models for simple domains, it remains laborious to do so for some complex real-world domains. Realistic planning action models are hard to encode, debug and maintain, and the development process is laborious. This presentation is pivoted on Machine Learning techniques which allow the learning of symbolic action models from a given set of traces consisting of Human-Robot interactions.

Bio:
Humbert Fiorino received his Ph.D. degree from SUPAERO (Ecole nationale supérieure de l'aéronautique et de l'espace) and the ONERA (Office National d'Etudes et Recherches Aérospatiales) in 1998. I also visited the Robotics and Artificial Intelligence group at the LAAS (Laboratoire d'Analyse et d'Architecture des Systèmes) as postdoctoral fellow. Since 1999, he is associate professor in computer science at Université Grenoble Alpes and permanent researcher at LIG (Laboratoire d’Informatique de Grenoble). His research works focus on Artificial Intelligence, automated planning and social robotics.

I will discuss a methodology for the generation of mobile robot controllers which offer probabilistic time-bounded guarantees on successful task completion, whilst also trying to satisfy soft goals. The approach is based on a stochastic model of the robot’s environment and action execution times, a set of soft goals, and a formal task specification in co-safe linear temporal logic, which are analysed using multi-objective model checking techniques for Markov decision processes. For efficiency, we propose a novel two-step approach. First, we explore policies on the Pareto front for minimising expected task execution time whilst optimising the achievement of soft goals. Then, we use this to prune a model with more detailed timing information, yielding a time-dependent policy for which more fine-grained probabilistic guarantees can be provided. We illustrate and evaluate the generation of policies on a delivery task in a care home scenario, where the robot also tries to engage in entertainment activities with the patients.

We present a family of logics for reasoning about agents' positions and movements in the plane which have several potential applications in the area of multi-agent systems, such as multi-agent planning and robotics. The most gen- eral logic includes (i) atomic formulas for representing the truth of a fact or the presence of an agent at a certain posi- tion of the plane, (ii) atomic programs corresponding to the four basic orientations in the plane (up, down, left, right) as well as the four program constructs of propositional dynamic logic PDL (sequential composition, nondeterministic compo- sition, iteration and test). As this logic is not computably enumerable, we study some interesting decidable and ax- iomatizable fragments of it. We also present a decidable extension of its iteration-free fragment by special programs representing movements of agents in the plane.

Energy games are two-player zero-sum games on a weighted graph, where one player want to maintain the accumulated weight within given bounds (possibly with an auxiliary objective). In this talk, I survey some results in the area.

In this paper, we present a robust system of self- directed autonomous robots evolving in a complex and public spaces and interacting with people. This system integrates high- level skills of environment modeling using knowlde-based mod- eling and reasoning and scene understanding with robust image and video analysis, distributed autonomous decision-making using Markov decision process and petri-net planning, short- term interacting with humans and robust and safe navigation in overcrowding spaces. This system has been deployed in a variety of public environments such as a shopping mall, a center of congress and in a lab to assist people and visitors. The results are very satisfying showing the effectiveness of the system and going beyond just a simple proof of concepts.

Abdel-Illah Mouaddib is a professor at the university of Caen Normandy and co-leader of GREYC-UMR6072 Lab. He received a PhD on Computer science on 1993. Professor Mouaddib is a recipient of the European Conference on AI (ECAI’98) Best Paper Award and nominated for the award in the IJCAI (International Joint Conference on Artificial Intelligence) 1999. His main research concerns topics of autonomous decision-making techniques of open systems in uncertain and partially observable environments and with interaction with humans.
He has been a publisher and a member of committees in the main conferences in the fields of artificial intelligence and robotics (IJCAI, AAAI, AAMAS, ICRA, ECAI, IROS). He served as an international expert of FWO, Netherlands, the USA NSF program and Canadian NSERC and member of committees of ANR. He was a collaborator of two NASA projects on Cooperative Robots (1999-2003), the PI of the projects COROCOP (control of cooperative robots) of ROBEA French program on Robotic and Artificial Entities (2004), and Robots_Malins of the ANR (2010-2012). He actively participated and leaded work packages in a lot of ANR projects (AMORCES’2007-2010, R_Discover’2008-2012, CRITERE’2010- 2012 and Lardons’2010-2012). He is actually the PI of the CHIST-ERA project COACHES and the DGA/ANR ATRID project GARDES.

Graph theory has a predominant role in the field of multi-robot coordination problems (spanning, e.g., distributed formation control and estimation schemes). Indeed, graphs are a convenient (combinatorial) abstraction for representing the various sensing/communication constraints among pairs of robots in the environments. Vertexes represent robots, and edges the possibility for a robot to measure/communicate with another robot in the group.
In this context, the notion of graph rigidity is gaining popularity since rigidity of a multi-robot "framework" (i.e., formation) has proven to be a necessary condition for the solution of many formation control/cooperative localization problems. Moreover, exploiting the tools of algebraic graph theory, one can associate combinatorial graph properties, such as rigidity, to some suitable spectral properties (i.e., eigenvalues) of associated matrixes. This spectral interpretation of graph-related properties makes it possible to ultimately design simple "gradient-like" controllers for the robot group able to solve formation control and localization problems in a decentralized way.
This talk will review the concepts of graph rigidity in the context of multi-robot applications, and present some recent results obtained with a team of quadrotor UAVs (drones) used as robotics platforms.

Games on graphs are useful models of multi-agent systems. If agents have local views of the world, then the corresponding games are of imperfect information. Unfortunately, very basic questions about such games are undecidable (for instance, strategy synthesis). In this talk I will advertise the class of games of imperfect information in which all actions are public. I will argue that this class strikes a good balance between computational complexity and modeling power.

Many modern robotic applications call for robots that can exhibit highly adaptive behaviours. For example, the development of domestic robotics requires a robot that can interact in unstructured environments with both human and non-human agents. To date, most approaches to building complex robotic systems have been largely ad-hoc; using software frameworks to plug together components that implement algorithms for vision, mapping, navigation, manipulation, and reasoning. In this talk I will introduce a framework for integrating disparate representations within a robotic system. The formal nature of this framework may allow us to establish meta-theoretic properties as well as to prove properties of a particular hierarchical instantiation. I argue that using more formal frameworks, such as the framework proposed in this talk, is a crucial component if we are to build adaptive, complex robots that can operate effectively in unstructured environments.
The talk is based on work by David Rajaratnam, Bernhard Hengst, and Michael Thielscher at UNSW, Sydney, Australia.
Bio: Abdallah Saffidine is an ARC DECRA Fellow and a Research Associate at the University of New South Wales, Sydney, Australia. He works in the Artificial Intelligence and Algorithms groups. He arrived at UNSW in 2013 as a postdoc with Pr. Michael Thielscher and obtained his PhD from the Universite Paris-Dauphine, France, under the supervision of Pr. Tristan Cazenave. Abdallah has a wide range of interests from games, planning, and other areas of decision-making to logic, complexity and other areas of theory.

This talk deals with the problem of active sensing control for nonlinear differentially flat systems. The objective is to improve the estimation accuracy of an observer by determining the inputs of the system that maximise the amount of information gathered by the outputs over a time horizon. In particular, we use the Costructibility Gramian (CG) to quantify the richness of the acquired information. First, we define a trajectory for the flat outputs of the system by using B-Spline curves. Then, we exploit an online gradient descent strategy to move the control points of the B-Spline in order to actively maximise the smallest eigenvalue of the CG over the whole planning horizon. While the system travels along its planned (optimized) trajectory, an Extended Kalman Filter (EKF) is used to estimate the system state and to keep memory of the past acquired sensory data for online re-planning. This is then used for an online replanning of the optimal trajectory during the robot motion which is continuously refined by exploiting the state estimation obtained by the EKF. In order to show the effectiveness of our method we consider the case of a unicycle that moves on a plane while measuring squared distances w.r.t. two markers whose positions are known. The simulation results show that, along the optimal path, the EKF converges faster and provides a more accurate and precise estimate than along other possible (non-optimal) paths.

In this talk, we will introduce dynamic epistemic logic that enables to represent and reason about higher-order knowledge (a knows that b knows that...). I will present basics about event models. Then I will make a summary of algorithms for model checking, satisfiability and epistemic planning.