Phd Position F/m Contact-centric Control of Robots - Strasbourg, France - Inria

Inria

Entreprise vérifiée

Strasbourg, France

il y a 2 semaines

Posté par:

Sophie Dupont

beBee Recruiter

Description

Le descriptif de l'offre ci-dessous est en Anglais_

Type de contrat :

CDD

Niveau de diplôme exigé :
Bac + 5 ou équivalent

Fonction :
Doctorant

Niveau d'expérience souhaité :
Jeune diplômé

Contexte et atouts du poste:

Context
The Artemis program aims to establish a sustained human presence on the Moon by the 2030s.

Space agencies are actively involved in designing and constructing a lunar base, which will serve as a habitat and operational hub; with this long-term vision, our team is collaborating with the CNES on providing a teleoperation approach for flexible teleoperation with large latencies.

The need for human labor to build, maintain, and perform tasks at the lunar base is limited, leading to the use of robotic systems.

While robots can handle routine tasks, many situations will require human supervision or control for adaptation and novel uses. For example, unforeseen maintenance operations may require a robot to use a tool, which was not preprogrammed.

Options for human involvement include on-site personnel or remote control from orbit, but these options do not fully address manpower constraints.

Therefore, the preferred solution is remote Earth-based robot control.

The primary challenge in Earth-to-Moon teleoperation is the 7-second round-trip delay, as observed in the NASA Lunar Laser Communication Demonstration (Boroson, With a sufficient bandwidth of 20 Mbps for HD video streaming, our team has experience in teleoperation with delays of up to 2 seconds using machine learning to predict future commands. However, we recognize that this technique will not scale to Earth-to-Moon delays.

Teleoperation, a classic topic in robotics, particularly in space robotics (Niemeyer, 2016; Pedersen 2003), offers three potential methods for Earth-to-Moon teleoperation:

Supervised autonomy, which involves remotely activating the robot's autonomous abilities, as seen in Mars rover operations (Squyres, 2005).
Direct teleoperation, where operators control the robot's arm(s) in realtime, as demonstrated in ESA experiments with an 800ms round trip delay (Panzirsch, 2023).
The use of a "predictive display," which involves operating a remote simulation and sending commands to the robot to execute without realtime feedback.

Co-Supervision
This PhD will be co-funded by CNES and co-supervised with Thomas Pietrzak (Inria Lille).

Mission confiée:

Objectives
Supervised teleoperation lacks flexibility as autonomous capabilities must be pre-designed. Direct teleoperation is flexible and intuitive for new tasks but struggles with long delays (>2s).

Predictive displays demand an accurate 3D representation of the remote environment and predicting action consequences, which is challenging, especially for outdoor Moon operations.

Our objective is to study an alternative approach for remote teleoperation that is (1) more flexible and intuitive than supervised teleoperation, (2) can cope with 5-10 seconds round-trip delays, (3) does not require predicting action consequences in an unfamiliar remote environment.

Approach
Our first hypothesis emphasizes the significance of contact in robotic operations, where user input is most valuable.

For example, in actions like pushing or grasping an object, the critical factor is finger placement, while the rest of the motion can be automatically planned by the robot's sensors.

Our second hypothesis posits that, while computers struggle to predict all outcomes in a novel environment, trained operators can anticipate consequences, often triggering actions.

Based on these ideas, we propose a novel teleoperation mode using contact points and 3D visualization from an RGB-D sensor like the MS Kinect.

Operators use a teleoperation device, such as two local arms, to select contact points in a static point cloud.

They guide the arm to the final position using the point cloud as a reference, without simulating physics, and initiate the action.

This approach is expected to be more flexible and intuitive than traditional supervised teleoperation, even with 5 to 10 seconds of delay.

We will use the Tiago++ robot from Inria, introducing artificial delays to simulate the Earth-to-Moon delay.

Our experienced team will conduct Human-Machine studies, comparing task completion time and success rates in comparison to direct and supervised teleoperation.

We will collaborate with Thomas Pietrzak from Inria Lille, who specializes in Human-Computer Interaction, particularly with non-traditional hardware.

References

Bicchi et al

Robotic grasping and contact:
A review. Proc ICRA 2000.

Boroson et al. Overview and results of the lunar laser communication demonstration. In Free-Space Laser Communication and Atmospheric Propagation XXVI, 201
Niemeyer et al. Telerobotics. Springer handbook of robotics, 201
Panzirsch et al. Exploring planet geology through forcefeedback telemanipulation from orbit. Science Robotics, 202
Pederse