Section: New Results

Navigation of Mobile Robots

Visual Navigation from an Image Memory

Participants : Suman Raj Bista, Paolo Robuffo Giordano, François Chaumette.

This study is concerned with visual autonomous navigation in indoor environments. As in our previous works concerning navigation outdoors [4], the approach is based on a topological localization of the current image with respect to a set of keyframe images, but the visual features used for this localization as well as for the visual servoing are not composed of points of interest, but either on mutual information [71] following the idea proposed in [3], or straight lines that are more common indoors [38], or finally on a combination of points of interest and straight lines [11]. Satisfactory experimental results have been obtained using the Pioneer mobile robot (see Section 6.9.2).

Robot-Human Interactions during Locomotion

Participant : Julien Pettré.

In collaboration with the Gepetto team of Laas in Toulouse and the Mimetic group in Rennes, we have studied how humans avoid collision with a robot. Understanding how humans achieve such avoidance is crucial to better anticipate humans' reactions to the presence of a robot and to control the robot to adapt its trajectory accordingly. It is generally assumed that humans avoid a robot just like they avoid another human. In this work, we bring the empirical demonstration that humans actually set a specific strategy to avoid robots, and that, more precisely, they show a preference to give way to a robot which is on a collision course with them [41]. This results brings useful insight about human-robot interactions during locomotion, and provides useful guidelines to design reactive navigation techniques for mobile robots aimed at moving among humans.

Scene Mapping based on Intelligent Human/Robot Interactions

Participant : Patrick Rives.

For mobile robots to operate in compliance with human presence, interpreting the impact of human activities and responding constructively is a challenging goal. Towards this objective, mapping an environment allows robots to be deployed in diverse workspaces, marking this skill as a primary element in the integration of robots into human-populated environments. We proposed an effective approach for using human activity cues in order to enhance robot mapping and navigation and in particular in filtering noisy human detections, detecting passages, inferring space occupancy and allowing navigation within unexplored areas. Our contributions [36] are based on the development of intelligent interactions among conceptually different mapping levels, namely, the metric, social and semantic levels. Experiments, using the Hannibal platform (see Section 6.9.2), highlighted a number of strong dependences among these levels and the way in which they can be used to enhance individual performances and in turn the global robot operation.

Autonomous Social Navigation of a Wheelchair

Participants : Vishnu Karakkat Narayanan, Marie Babel.

This work is realized in collaboration with Anne Spalanzani (Chroma team - Inria Grenoble).

A key issue that hinders the adoption of assistive robotic technologies such as robotized wheelchair, in the real world, is that they need to operate in mostly human environments and among human crowds. Indeed intelligent wheelchairs need to be deployed in a human environment thereby making it essential for such robots to incorporate a sense of human-awareness. Simply put, humans are special objects that have to be perceived and acted on in a special manner by robots that interact with us humans. Thus one can define Human-aware Navigation as an intersection between human-robot interaction and robotic motion planning.

In this context we introduced a low-level velocity controller that could be employed by a social robot like a robotic wheelchair for approaching a group of interacting humans, in order to become a part of the interaction. Taking into account an interaction space that is created when at least two humans interact, a meeting point can be calculated where the robot should reach in order to equitably share space among the interacting group. We then proposed a sensor-based control law which uses the position and orientation of the humans with respect to the sensor as inputs, to reach the meeting point while respecting spatial social constraints [61]. Experiments using a mobile robot equipped with a single laser scanner, realized in collaboration with Ren Luo (Taiwan) within the Sampen Inria associated team, also proved the success of the algorithm in a noisy real world scenario [62].

In addition, a semi-autonomous framework for human-aware navigation in an intelligent wheelchair has been designed. A generalized linear control sharing framework was proposed that was able to progressively correct the user teleoperation in order to avoid obstacles and in order to avoid disturbance to humans. Meanwhile, we also proposed a Bayesian approach for user intention estimation. The formulation was partly inferred from the design of the controller for assisted doorway passing, wherein we hypothesized that predicting short term goals is sufficient for eliminating user intention uncertainty [16].

Semi-autonomous Control of a Wheelchair for Navigation Assistance

Participants : Louise Devigne, Vishnu Karakkat Narayanan, Marie Babel.

To address the wheelchair driving assistance issue, we proposed a unified shared control framework able to smoothly correct the trajectory of the electrical wheelchair [16]. The system integrates the manual control with sensor-based constraints by means of a dedicated optimization strategy. The resulting low-complex and low-cost embedded system is easily plugged onto on-the-shelf wheelchairs [55]. The robotic solution is currently under validation process with volunteering patients of Pôle Saint Hélier (France) who present different disabling neuro-pathologies preventing them to drive non-assisted wheelchairs.

Within the frame of ISI4NAVE associated team (see Section 9.4.1.2), this shared-control solution has been then coupled with first experimental biofeedback devices such as haptic devices. Preliminary tests have been conducted within the PAMELA facility at University College of London and within the rehabilitation center of Pôle Saint Hélier in Rennes (see Section 8.1.3). They involved regular wheelchair users as well as medical staff. We have demonstrated the ability of the framework to provide relevant assistance and now need to focus on methods to fine-tune parameters and customize/calibrate to the individual and evolving needs of each user.