The COVID-19 pandemic has spurred a large amount of experimental and observational studies reporting clear correlation between the risk of developing severe COVID-19 (or dying from it) and whether the individual is male or female. This paper is an attempt to explain the supposed male vulnerability to COVID-19 using a causal approach. We proceed by identifying a set of confounding and mediating factors, based on the review of epidemiological literature and analysis of sex-dis-aggregated data. Those factors are then taken into consideration to produce explainable and fair prediction and decision models from observational data. The paper outlines how non-causal models can motivate discriminatory policies such as biased allocation of the limited resources in intensive care units (ICUs). The objective is to anticipate and avoid disparate impact and discrimination, by considering causal knowledge and causal-based techniques to compliment the collection and analysis of observational big-data. The hope is to contribute to more careful use of health related information access systems for developing fair and robust predictive models.

We address ethical dilemma situations that may arise during autonomous driving. To evaluate how this deliberation could work, we propose a decision-making algorithm based on a Markov Decision Process (MDP) which controls the vehicle in normal conditions. When a dilemma situation is detected, the collision severity is determined by an evaluation of the harm incurred by different types of road users. Then, to illustrate different moral approaches, three different policies are proposed: one based on ralwsian contractarianism, another on utilitarianism, and finally on egalitarianism. Each policy supports a different view of the concept of fairness, potentially producing different behaviors for the same dilemma situation.

An important current challenge in Human-Robot Interaction (HRI) is to enable robots to learn on-the-fly from human feedback. However, humans show a great variability in the way they reward robots. We propose to address this issue by enabling the robot to combine different learning strategies, namely model-based (MB) and model-free (MF) reinforcement learning. We simulate two HRI scenarios: a simple task where the human congratulates the robot for putting the right cubes in the right boxes, and a more complicated version of this task where cubes have to be placed in a specific order. We show that our existing MB-MF coordination algorithm previously tested in robot navigation works well here without retuning parameters. It leads to the maximal performance while producing the same minimal computational cost as MF alone. Moreover, the algorithm gives a robust performance no matter the variability of the simulated human feedback, while each strategy alone is impacted by this variability. Overall, the results suggest a promising way to promote robot learning flexibility when facing variable human feedback.

Codes of ethics provide a critical foundation for an organization to define the tenets that best reflect their values. These codes provide a touchstone for employees to follow that establishes trust between an organization and their customers and key stakeholders. The rapid acceleration of autonomous and intelligent systems (A/IS) has created an imperative for technologists to complement these codes with new methodologies focused on the widespread implementation of ethical considerations in all levels of design and manufacture. This paper describes IEEE's work in this arena, including the IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems, the IEEE TechEthics program, the IEEE P7000 Standards Series focused on A/IS issues, Digital Inclusion through Trust and Agency, and the ongoing work of the IEEE Society on Social Implications of Technology. This paper also includes a table of approximately 50 activities within the IEEE related to ethics.

Ethics by Design concerns the methods, algorithms and tools needed to endow autonomous agents with the capability to reason about the ethical aspects of their decisions, and the methods, tools and formalisms to guarantee that an agent's behavior remains within given moral bounds. In this context some questions arise: How and to what extent can agents understand the social reality in which they operate, and the other intelligences (AI, animals and humans) with which they co-exist? What are the ethical concerns in the emerging new forms of society, and how do we ensure the human dimension is upheld in interactions and decisions by autonomous agents?. But overall, the central question is: "Can we, and should we, build ethically-aware agents?" This paper presents initial conclusions from the thematic day of the same name held at PRIMA2017, on October 2017.

As humanoid robots start entering to the human beings environments, the collision between the robot and other objects is inevitable. To solve this problem, we propose the human mimicking push-recovery strategy. With the arm mechanism, the humanoid robot utilizes human-like strategies to deal with the unexpected collision. We analyse the human push-recovery movement to the external force. Thus we can design the trajectory generator to mimic human beings reactions. To reduce the modelling error and improve zmp stability, we utilizes the five-mass with angular momentum model as humanoid robot model. We design the safe bound examination and the Center of Mass (COM) state estimator to judge the collision stage. The safe bound examination is used to guarantee the walking stability. Moreover, the COM state estimator is used to determine whether the pushing is over or not. The push-recovery control system is implemented on the humanoid robot developed in our NTU-iCeiRA lab. The purpose of this paper aims to integrate different push-recovery strategies while encountering different collision conditions based on the five-mass with angular momentum model.

Considering perception as an observation process only is the very reason for which robotic perception methods are to date unable to provide a general capacity of scene understanding. Related work in neuroscience has shown that there is a strong relationship between perception and action. We believe that considering perception in relation to action requires to interpret the scene in terms of the agent's own potential capabilities. In this paper, we propose a Bayesian approach for learning sensorimotor representations through the interaction between action and observation capabilities. We represent the notion of affordance as a probabilistic relation between three elements: objects, actions and effects. Experiments for affordances discovery were performed on a real robotic platform in an unsupervised way assuming a limited set of innate capabilities. Results show dependency relations that connect the three elements in a common frame: affordances. The increasing number of interactions and observations results in a Bayesian network that captures the relationships between them. The learned representation can be used for prediction tasks.

Robot navigation in human environments has been in the eyes of researchers for the last few years. Robots operating under these circumstances have to take human awareness into consideration for safety and acceptance reasons. Nonetheless, navigation have been often treated as going towards a goal point or avoiding people, without considering the robot engaging a person or a group of people in order to interact with them. This paper presents two navigation approaches based on the use of inverse reinforcement learning (IRL) from exemplar situations. This allow us to implement two path planners that take into account social norms for navigation towards isolated people. For the first planner, we learn an appropriate way to approach a person in an open area without static obstacles, this information is used to generate robot's path plan. As for the second planner, we learn the weights of a linear combination of continuous functions that we use to generate a costmap for the approach-behavior. This costmap is then combined with others, e.g. a costmap with higher cost around obstacles, and finally a path is generated with Dijkstra's algorithm.

This paper introduces two novel algorithms for detecting groups of people standing or freely moving in a crowded environment. The proposed algorithms exploit low-level features extracted from videos. The first algorithm, the Link Method, uses a learning and forgetting strategy for modeling dynamics of proxemics between individuals. Two versions of this algorithm are proposed: they differ in the analysis of proxemics. The second one, called Interpersonal Synchrony Method, explicitly adopts interpersonal synchrony to refine clusters of persons detected by combining together proxemics and 2D field of view of individuals. The algorithms are evaluated on both simulated and real-world video sequences from state-of-the-art databases. Clustering metrics such as the Adjusted Mutual Information shows that our models outperform the approach based on F-formations. This work developed algorithms that can be readily applied in robotics, to allow robots to automatically detect groups in crowded environments.

In order to define an architecture for task and motion planning of a mobile robot, we propose the Cell-RRT path planner that combines the advantages of planning approaches by decomposition of the environment and the advantages of probabilistic approaches. Experiments of the method for various decomposition granularities and various adjustments of the planner settings show that using a bias towards the goal while choosing a random configuration reduces the paths length but can cause failures, that the choice of the criterion for analysing the environment is important, and that the method can profit from a reuse of already made computations in a part of the environment.

If robots have to be, one day, part of our environment and assist humans in their daily life, they will have to be endowed not only with the necessary functions for sensing, moving and acting, but also and inevitably, with more advanced cognitive capacities. Indeed a robot that will interact with people will need to be able to understand the spatial and dynamic structure of its environment, to exhibit a social behavior, to communicate with humans at the appropriate level of abstraction, to focus its attention and to take decisions for achieving tasks, to learn new knowledge and to evolve new capacities in an open-ended fashion. The COGNIRON (The Cognitive Robot Companion) project1 studies the development of robots whose ultimate task would be serve and assist humans. Research is carried out in four main domains: Perception and spatial cognition. The project addresses the fundamental questions of scene understanding, object learning and recognition, and semantic labeling of space through interacting with humans. Human-Robot interaction. Enabling an embodied dialogue capability is essential for a cognitive robot companion interacting with humans. One of the crucial aspects is handling multi-modality to cope gestures and other non-verbal signals and to make references to the environment shared with humans. Another aspect of Human-robot interaction lies in understanding human actions by the robot so that it can act in an appropriate manner and understand human needs. This requires visual detection and tracking of humans, of their faces, 3D modeling of human body, interpretation of human activities based on gestures, postures, attitudes, and motions. Finally, the robot must exhibit an acceptable social behavior, respecting human social spaces and affective factors (e.g., user comfort). Decision-making and expression of intentions. Making decisions, be it for autonomous task achievement or for humanrobot interaction, is a fundamental capability of a cognitive robot. The project studies decision-making abilities in uncertain and varying environments, shared decision-making and actions with humans, intentionality expression by robots during interaction with humans, as well as cognitive architectures for the integration of robot capabilities. Learning. For a robot companion to show adaptive, life-long learning behavior, it must be capable of acquiring new skills when required, and of incremental acquisition of skills for learning complete tasks. Imitation learning from humans is one of the main approaches investigated by the project to achieve this learning capability. The talk will overview the achievements of the project and its results that are demonstrated in three main experimental settings, enabling to exhibit the abovementioned cognitive capacities : the robot home tour, the curious and proactive robot and the robot learner.

Some recent stereo matching algorithms are based on graph cuts. They transform the matching problem to a minimisation of a global energy function. The minimisation can be done by finding out an optimal cut in a special graph. Different methods were proposed to construct the graph. But all of them, consider for each pixel, all possible disparities between minimum and maximum values. In this article, a new method is proposed: only some potential values in the disparity range are selected for each pixel. These values can be found using a local analysis of stereo matching. This method allows us to make wider the disparity range, and at the same time to limit the volume of the graph, and therefore to reduce the computation time.

Measuring the performance of robots, and more precisely of autonomous robots dealing with complex and changing environment, is a long standing issue. There is today no satisfactory known metric to measure quantitatively and precisely how a robot system satisfies accurately the specification expectations, neither do we know how to qualify and delineate in general the operational domains of the robot. This paper presents an approach aiming at qualifying robot autonomy. We present a method based on the measure of robot performance in achieving a given task related to environment complexity. In order to test a large number of tasks and environment configurations, we have designed a robot simulator that allows the use of the same algorithms implemented on the real robot and in simulation. To illustrate the work, we present results based on hundreds of simulations of a navigation mission and compare robot performance when algorithm parameters change

Determining the performance of a robot is a challenge because success or failure depends not only on the capabilities of the robot but on the difficulties of the environment as well. This paper presents a method of understanding the performance of a robot with respect to its environment through the interaction of a set of metrics. Metric interactions are learned in a dynamic Bayesian network and placed in a probabilistic systems model called a system map, which is used to understand how the metrics relate both to each other and to a partially known environment. Initial results presented here demonstrate how this model identifies environmental dependencies and how performance can be predicted even in an uncertain environment

Rackham is an interactive robot-guide that has been used in several places and exhibitions. This paper presents its design and reports on results that have been obtained after its deployment in a permanent exhibition. The project is conducted so as to incrementally enhance the robot functional and decisional capabilities based on the observation of the interaction between the public and the robot. Besides robustness and efficiency in the robot navigation abilities in a dynamic environment, our focus was to develop and test a methodology to integrate human-robot interaction abilities in a systematic way. We first present the robot and some of its key design issues. Then, we discuss a number of lessons that we have drawn from its use in interaction with the public and how that will serve to refine our design choices and to enhance robot efficiency and acceptability

In this article, we focus on environment models for the well-known simultaneous localisation and map building (SLAM) problem, which has received considerable attention in the robotics community over the past few years. First, we compare different existing map representations to discuss their advantages and limitations in the scope of indoor robotics applications. Then, we define a highly structured map model which combines different kinds of representations, including space-based, grid-based as well as feature-based formats. This model also provides topological information such as adjacency links, which are similar to the topological layer used in geographical information systems. We explain how to build and update maps according to this model, using a mobile robot equipped with a laser scanner and underline how the structure of our representation may increase robustness in a Kalman-based SLAM process. Finally, we show some preliminary experiments and propose a few perspectives for this work.

Most robotic systems are designed for given goals. Even learning systems follow this paradigm by trying to improve overall performance for a given task. These systems are limited by the knowledge of the developers and are not able to overpass their initial set of goals. We propose to explore a new kind of sensory motor systems that are able to acquire new representations and new goals starting from an initial small set. Instead of developing algorithms for a given task, we want to develop a general approach that is task acquisition oriented. The work reported in this paper is just a beginning and propose a theoretical framework and first results for such systems.

In this article we propose a method for landmark selection and tracking in natural environments. Landmarks are not specific 3D objects, but an image area that can be tracked over a sequence of monocular images. No assumption is made a priori on landmarks. The algorithm selects areas in a monocular black and white image sequence that have an affine trackability property. A measure of affine trackability is proposed for selection and affine tracking. Initial experimental results are presented.

Autonomous robots are complex machines embedding: software modules implementing basic functions such as servo loops to follow a path, or to compute stereo correlation of a pair of images; software to take into account the interactions between the robot components; programs which deal with the supervision and the planning of the mission. These components are integrated within an architecture which defines an organization and a methodology for robot operation.

The authors previously (1998) argued that the LAAS architecture is one of the most suitable for mobile robot control. This statement may seem over-optimistic, not to say pretentious and unverifiable. After all, can we compare architectures? can we set up benchmarks? or can we measure how good an architecture is compared to another? An architecture defines organization principles, integration methods and supporting tools. Comparing those tools, methods and principles may sometime end up in sterile controversies. However, we think there are means to measure the overall quality (or interest) of an architecture. Development time is for example one relevant criterion. Basically, using a specific architecture, how long does it take to integrate a complete demonstration, including nontrivial decisional capabilities, from the low level functional modules up to the supervisory level? This may seem a rather weak measure of architecture quality; however, it encompasses properties such as genericity and adaptability, ease of design and programming, extensibility and robustness. In this paper we describe our recent experience in integrating a complete demonstration from scratch in 40 days using the LAAS architecture.

The paper presents an approach to reactive navigation in cross-country terrains. The approach relies on a particular probabilistic obstacle detection procedure, that describes the area perceived by a pair of stereo cameras as a set of polygonal cells. To generate the motion commands on the basis of this terrain description, we present some improvements and adaptations to the classical potential fields technique. Results on real stereo data illustrate our contribution throughout the paper, and simulated long range traverses are discussed.

In aim to optimize the control for an off-road robot and specially a robot with a crawling motion mode named peristaltical motion, it is essential to understand robot-soil interaction. We present an efficient method for simulating this interaction. This method can simulate the dynamic behavior of a 6-wheeled/3-axles Marsokhod type robot on different kinds of soils. With this simulation, we have defined the necessary conditions for using the peristaltical motion on sand or loose soil.

This paper presents on operational framework to bridge the gap between planning with uncertainty and real-time sensor-based motion control. The environment being known, a planner produces a plan composed of free space and sensor-based motion commands. The representations of uncertainty and its evolution, environment landmarks, and actions generated at the planning level are discussed. Sensor-based actions and command definitions for a nonholonomic mobile robot based on a task-potential field approach are developed. These various elements are integrated in a system that actually generates the motions of the Hilare2 mobile robot.

A method combining planning and reactive control for car-like nonholonomic mobile robots is discussed. Firstly, a "bubble" for a car-like mobile robot is defined as the locally reachable space from a given configuration considering the obstacles and using the appropriate metric. Then a flexible feasible trajectory, based on the elastic band concepts, is constructed. This trajectory is smoothed using Bezier curves satisfying a minimum curvature constraint, and a parameterization is proposed which satisfies the robot kinematics constraints.

This paper presents a general methodology for the specification and the integration of functional modules in a distributed reactive robot architecture. The approach is based on a hybrid architecture basically composed of two levels: a lower distributed functional level controlled by a centralized decisional level. Due to this methodology, synchronous or asynchronous operating capabilities (servo-control, data processing, event monitoring) can be easily added to the functional level. They are encapsulated into modules, built according to a generic model, that are seen by the decisional level as homogeneous, programmable, reactive and robust communicant services. Each module is simply described with a specific language and is automatically produced by a generator of modules (G/sup en/oM) according to the generic model. G/sup en/oM also produces an interactive test program and interface libraries to control the module and to read the resulting data, which allow one to directly integrate the module into the architecture.

Building on previous work on incremental natural scene modelling for mobile robot navigation, we focus in this paper on the problem of representing and managing uncertainties. The environment is composed of ground regions and objects. Objects (e.g., rocks) are represented by an uncertain state vector (location) and a variance-covariance matrix. Their shapes are approximated by ellipsoids. Landmarks are defined as objects with specific properties (discrimination, accuracy) that permit to use them for robot localization and for anchoring the environment model. Model updating is based on an extended Kalman filter. Experimental results are given that show the construction of a consistent model over tens of meters.

A mobile robot navigates in an environment and incrementally builds a map while locating itself. The quality of the map depends on the sensor models. Due to the difficulty of defining a stochastic odometric model, the classical approach based on Kalman filtering is revisited in order to decouple the information provided by the odometric sensor and exteroceptive sensors. The proposed approach is based on the observation and fusion of geometric relationships between objects. These relationships are represented in local maps and related in a global map where odometric information is exploited. The resulting local maps are invariant with respect to biases introduced in the robot location (position and orientation) due to sliding for instance. Experimental results on real data compare both the classical and the proposed approach.

We discuss Procedural Reasoning System (PRS) as a high-level control and supervision language adapted to autonomous robots to represent and execute procedures, scripts and plans in dynamic environments. We discuss the main reasons why PRS is well suited for this type of application: (1) The semantics of its plan (procedure) representation, which is important for plan execution and goal refinement; (2) Its ability to construct and act on partial (rather than complete) plans; (3) Its ability to pursue goal-directed tasks while at the same time being responsive to changing patterns of events in bounded time; (4) Its facilities for managing multiple tasks in real-time; (5) Its default mechanisms for handling stringent real-time demands of its environment; and (6) Its meta-level (or reflective) reasoning capabilities. C-PRS has been used to implement an embedded control and supervision system for autonomous mobile robots in two different experimentations that we briefly present. We conclude with some suggestions to further develop C-PRS and with a short review of related work.

We describe the specification of a software control architecture for autonomous mobile robots. The architecture, designed to provide the robot (in a task-dependent context) with the capacity to react to events but also to intelligently anticipate the future and plan its actions, is based on the decomposition of the robot system into a functional and a decisional level. The article is mainly focused on some aspects of the organisation and of the operation of the system such as execution control, inter-levels communication, reactivity. An important aspect that is developed is the possibility to prove some temporal and logical properties of parts of the system.

Mobile robot navigation in natural environments requires incremental scene modelling and robot self-localization. These two capacities are discussed in this paper. A scene model is built with heterogeneous representations to match the diversity of the environment. We focus on object-based representations, using coarse models (ellipsoids) or volumic primitives (superquadrics), and their topological relations. The method for extracting the models from a range image to provide geometrical and topological representations, and the matching of successive views during the robot's movement are presented. For robot localization, two procedures are proposed based on the coarse or fine models. Experimental results are given.

Clothoids are very useful for smoothing the motion of a mobile robot moving along a trajectory. This paper addresses the problem of smoothing mobile robot motions when cusps, i.e., changes of motion direction along the trajectory, are imposed. We pinpoint some special curves (that we call "anticlothoids") and we discuss how they can be used together with clothoids in order to smooth a predefined trajectory.>

This paper deals with the architecture and control structure of mobile robots. The authors decompose robot functions into modules organized according to their predefined interactions: sensor modules that accomplish various processings on data from physical sensors, effector modules that issue commands to effectors, servo-processes that establish links between perception and action to achieve closed-loop behaviors, and functional units that provide specific functionalities. These modules, and hence the robot system itself, are controlled by a control system that also enables the robot to execute missions (plans) expressed in a command language. The authors introduce and discuss a generic control system structure, composed of a supervisor that interprets the plan and oversees its execution, an executive for operating and managing robot modules and resources, a surveillance manager for detecting and reacting to asynchronous events, and an error recovery module for local plan mending and correction. Several experimental examples are given.>

In this paper, we focus on some perceptual functions required by a generic task "GOTO" in natural environments; in previous works, only geometrical modeling has been used to deal with two fundamental tasks: landmark extraction and recognition for sensor-based motion control or robot localization, and terrain modeling for motion planning. Geometrical representations alone lead to a bulky model, and, after some iterations, to a combinatorial explosion. We present here, higher level representations; from a range image, given some assumptions on the perceived scene (even ground with few objects), we propose a segmentation algorithm to extract simple semantical representations for the ground and the objects; then, we can analyze the relative positions of the objects to build a topological scene description. Both models constitute the scene model, needed for further incremental environment modeling.>

This paper proposes a formalism for the description and automated implementation of the functional modules of an autonomous mobile robot in the framework of multi-layered control architectures. The formalization of the description of a module, including its behavior and its interfaces with other modules makes the implementation easier by providing higher level tools and automatic consistence checking tools in the module compiler. The execution control level can take advantage of the formal description of the activity tree of the modules to manage the resources more efficiently. A global task of the robot is achieved by the control system by connecting the modules into a dynamic activity tree. Finally the authors present a modules compiler that uses the formal module description and user supplied functions to produce the modules in an automated way.>

This paper presents the approach, algorithms and processes we developed to perform cross-country autonomous navigation. After a presentation of the teleprogramming context, we introduce an adaptive navigation approach, well suited for the characteristics of complex natural environments. The main perception, motion planning and decisional processes required by the robot during navigation are briefly presented. An on board control architecture that manages all these processes is then described, and first results of an experiment currently developed at LAAS are discussed.>

Considers the problem of programming a remote intervention robot with a high degree of autonomy and reasoning abilities. The authors propose a control strategy, called task-level teleprogramming, based on a specific architecture. After a brief presentation of their approach, the authors discuss in this paper developments, mainly a teleprogramming language, and illustrate them with an example.>

The authors consider the class of intervention robots that have to perform tasks in remote environments which may impose painful, harsh or utterly dangerous conditions to humans. They informally characterize this class of problems and briefly discuss several basic concepts which underlie the current approaches. The remote autonomous robot concept that they have developed fulfills the requirements deduced of these basic ideas. The authors illustrate the real world possibilities of the concept by describing the features of two application projects (AMR, VAP) that they believe to be significant and demonstrative. The last section is devoted to the description of a generic experiment carried on with ADAM, an all-terrain full size rover. It includes the system architecture implementation and some of the experimental results.

The authors describe an implemented integrated system allowing a mobile robot to plan its actions, taking into account temporal constraints, and to control their execution in real time. The general architecture has three levels, and the approach is related to hierarchical planning: the plan produced by the temporal planner is further refined at the control level, which in turn supervises its execution by a functional level. The framework of the French Mars Rover project VAP is used as an illustration of the various aspects discussed.>

We consider the class of intervention mobile robots that huve to perform tasks in ill-known environments which are often remote or of difflicult or dangerous access. An outstanding example of this would be a Mars Rover designed to carry out planet exploration, scientific measurements and experiments, according lo the requeslis of Earth based scientists over a long period of time, e.g. one Martian year. The lack of structure in the environment of this class of intervention robots, as well as the nonrepetitive nature of the tasks, clearly require eficient task programming and constant adaptation to the environment by the robot. We contend that this class of problems demands a specific decisional architecture which couples “intelligence” at the programming level with “intelligence” at the level of the physical machine. This second requirement implies that the machine exhibits the attributes of an autonomous intelligent robot. In this paper we develop the concepts of a generic architecture aimed to achieve task control of a remote intervention robot. The robot is fully autonomous at task level. It receives tasks assignements that it transforms into sequences of actions using ats own interpretation and planning capacities, and executes these actions while being ,reactive to asynchronous events and environment conditions. We have developed the approach presented here in the context of several domain-oriented projects, and fully implemented a first version of the concepts in our newest mobile robot, HILARE II. The framework of the french Mars Rover project VAP is used as an illustration of some crucial aspects discussed in the paper.

Explicit representation of free-space is necessary for mobile robot navigation. This paper presents a fully implemented system that achieves incremental environment modelling for a 2D mobile robot using a laser range-finder. This system produces a surfacic model of objects (polylines and associated uncertainties) and a model of free-space (polygons) taking into account possible appearance and disappearance of objects.>

The authors present a control system architecture for mobile robots able to execute missions specified by a user and having reactivity and error recovery capacities. The architecture is based on three types of entities: modules that accomplish basic computations on data from sensors; processes that establish dynamic links between perception and action to achieve closed-loop behavior; and functional units providing specific functionalities. These modules are organized hierarchically and are considered to be the robot's resources. The authors introduce and discuss a generic control system structure, which is based on a decomposition of its functions mainly into an executive for managing robot resources and a surveillance manager for detecting and reacting to asynchronous events. The control system enables the robot to execute missions (plans) expressed in a command language. Several running examples are given.>

In order to understand its environment, a mobile robot should be able to model consistently this environment, and to locate itself correctly. One major difficulty to be solved is the inaccuracies introduced by the sensors. The approach proposed in this paper to cope with this problem relies on 1) defining general principles to deal with uncertainties : the use of a multisensory system, favo ring of the data collected by the more accurate sensor in a given situation, averaging of different but consistent measurements of the same entity weighted with their associated uncertainties, and 2) a methodology enabling a mobile robot to define its own reference landmarks while exploring its environment. These ideas are presented together with an example of their application on the mobile robot HILARE.