High precision robotic manipulation with reinforcement learning and Sim2Real

High precision robotic assembly that handles high product variability is a key part of an agile and a flexible manufacturing automation system. To date however, most of the existing systems are difficult to scale with product variability since they need precise models of the environment dynamics in order to be efficient. This information is not always easy to get.
Reinforcement learning based methods can be of interest in this situation. They do not rely on the environment dynamics and only need sample data from the system to learn a new manipulation skill. The main caveat is the efficiency of the data generation process.
In this post-doc, we propose to investigate the use of reinforcement learning based algorithms to solve high precision robotic assembly tasks. To handle the problem of sample generation we leverage the use of simulators and adopt a sim2real approach. The goal is to build a system than can solve tasks such as those proposed in the World Robot Challenge and tasks that the CEA’s industrial partners will provide.

Conversational Agent for Medical Serious Games

The LVIC laboratory participates in a research project which aims to develop innovative tools for teaching medical students. The training will be done through serious games of second generation, in which the learner can interact directly with the environment:
- immersed in a 3D environment with a Virtual Reality Head Mounted Display and motion detection,
- with natural and ecological handling of the environment (instruments, patient …),
- and a voice interaction with conversational and emotional avatars.

The multimedia team of LVIC laboratory is involved in the project to develop tools allowing students to interact in natural language with conversational avatars.

In this context, the post-doctoral researcher will be in charge of:
- studying the state of art of conversational agents;
- understanding and mastering the technological components of the laboratory language processing;
- proposing and developing a dialogue system allowing interaction in natural language with conversational avatars of the project.

Simultaneous Localisation and Mapping with an RGB-D camera based on a direct and sparse method

Recent advances in the methods of locating a device (smartphone, robot) in relation
to its environment make it possible to consider the deployment of augmented reality solutions and autonomous robots. The interest of RGB-D cameras in such a context is notable since it allows to directly acquire the depth map of the perceived scene.
The objective of this post docorate consists in developping a new SLAM (Simultaneous Localisation and Mapping) method relying on a depth sensor.

To reach a solution both robust, accurate and with small CPU/memory comsumption, the depth image will be exploited though a direct and sparse approach. The resulting solution will be then combined with the solution of "RGB SLAM Constrained to a CAD model" developped in our laboratory, resulting finaly in an "RGB-D SLAM Constrained to a CAD model"

Large scale visual recognition

This post-doc deals with detection and recognition of objects in images and video streams, on a large scale. This is a fundamental task that is the subject of active research in the world, including recent challenges in the evaluation campaigns. The "large scale" aspect refer to both large size databases (eg ten million images) and large number of concepts to recognize (eg 100-10000).The work will concern bothimage description and classification.

At the description level, state of the art techniques rely on local descriptors, aggregated according to dictionaries of "visual words" possibly constructed using Fisher kernels. It is nevertheless necessary to recode these signatures effectively in order to manage large databases. Regarding learning of visual concepts or objects, many algorithms use SVM (support vector machines) but other approaches are sometimes considered, such as those based on boosting or logistic regression.

The proposed position involves research and development of efficient algorithms to find visual entities in very large databases. Tracks are considered and should be discussed with the candidate selected based on prior knowledge and technical discussions.