Post-doctoral fellow (M/F): Unsupervised learning of object manipulation policies for efficient learning of sensorimotor representations
Postdoc @French National Centre for Scientific Research posted 22 hours agoJob Description
General information
Job title: Postdoctoral fellow (M/F): Unsupervised learning of object manipulation policies for efficient learning of sensorimotor representations
Reference: UMR6602-CELTEU-001
Number of positions: 1
Work location: AUBIERE
Publication date: Friday, June 6, 2025
Type of contract: Researcher on fixed-term contract
Contract duration: 18 months
Expected hiring date: October 1, 2025
Workload: Full
Remuneration: from €2,991 gross monthly depending on experience
Desired level of studies: Doctorate
Desired experience: Indifferent
CN section(s): 07 – Information sciences: processing, integrated hardware-software systems, robots, controls, images, content, interactions, signals and languages
Missions
A first part of the MeSMRise project will focus on learning (multimodal) representations and interaction graphs structured by actions. While random or naive action policies can be used for this, directed manipulations should be more effective for learning object representations.
The candidate will therefore focus on learning action policies and will address the following main questions: – How to learn in an unsupervised way to select actions that will lead to better representations? We will consider the framework of learning active object manipulation and explore the use of intrinsic impulses derived from SSL losses to learn manipulation policies. – How to learn hierarchical policies for object manipulation using SSL losses as active learning engines? We will study the impact of accessing these different levels of actions in a hierarchical policy case. – How to exploit learned expectation graphs to guide the learning of efficient policies? We will also investigate how to best leverage the inference information provided by anticipations learned from other WPs to further guide the agent’s learning. Indeed, abstract representations (e.g., sensorimotor primitives) and higher-level inferences in non-Markovian environments can be used for optimal action planning (e.g., using the heuristic-informed search algorithm D*), structure augmentations (e.g., to link multiple small rotations as a larger-scale manipulation of a single object). This can be used to select the best course of action, in order to optimize exploitation (e.g., for distinguishability) or exploration (e.g., by biasing curiosity mechanisms). Finally, the candidate will be expected to contribute to the coordination with other project tasks and partners.
Activities
– Propose different policy learning strategies and architectures
– Evaluate these strategies using a 3D robotic environment simulating the manipulation of 3D objects
– Write scientific articles
– Coordinate with other project partners and contribute to integration.
SKILLS
The ideal candidate holds a PhD in a relevant field and has:
– a strong background and publications in the field of machine learning, in particular deep learning and reinforcement learning for object manipulation and perception.
– Experience with active learning, intrinsic motivations and/or self-supervised learning is highly desired.
– Experience with 3D robotics simulators
– Ability to interact smoothly with different members of the consortium;
– Autonomy and proactivity in research activities and activity reporting
Work context
This postdoctoral position is part of the ANR MeSMRise (Multimodal deep SensoriMotor Representation learning) project (https://projet.liris.cnrs.fr/mesmrise/index.html).
The MeSMRise project proposes to draw inspiration from how human babies learn to explore their environment through actions that shape their multimodal experience. Inspired by the theory of sensorimotor contingencies (SMC), the main objective of the project is to study how action can structure multimodal representations, learned with self-supervised learning (SSL) methods. This will be applied to 3D objects, perceived by vision and point clouds, and manipulated in virtual environments.
This postdoctoral position is part of the third workpackage of the project related to active learning, focusing on learning action policies that enable efficient learning of object representations.
The candidate will work at the Pascal Institute, near Clermont-Ferrand, and will interact with other project partners in Lyon and Grenoble.
The position is located in a sector covered by the Protection of Scientific and Technical Potential (PPST), and therefore requires, in accordance with regulations, that your arrival be authorized by the competent MESR authority.
Constraints and risks
Nothing to report
