Vision Engineer - Learning-Based Robotic Manipulation (F/M/D)

Linkedin; Lazio; In loco; EUR 30.000 - 36.000; Tempo pieno

Descrizione del lavoro

AICAM Mondovi Site in Rome, Italy is seeking a Vision Engineer to advance learning-based robotic manipulation from perception to real hardware deployment. You will work on pipelines that convert multi-camera 3D perception into actionable policies and robust bimanual manipulation in dynamic hospitality settings.

The role emphasizes building sim-to-real capable systems, training policies, and integrating with perception and control teams.

Competenze

• Experience with learning-based robotic manipulation on real hardware.
• Solid computer vision and 3D learning: point clouds, depth, multi-view geometry, and 6-DoF pose estimation or grasp regression.
• Hands-on experience with robotic grasping and/or imitation learning.
• Experience with teleoperation and data collection for robot learning.
• Proficiency in Python and PyTorch, with ROS 2 familiarity.
• Working experience with at least one robotics simulator (Isaac Sim / Lab, MuJoCo, ManiSkill).
• Proven ability to debug and tune robotic systems in real-world conditions, dealing with sensor noise, latency, and mechanical limitations.

Mansioni

• Own the teleoperation and demonstration data pipeline: the collection setup, data quality and curation, and the path from recorded episodes to trained policies.
• Build high-fidelity simulation of the robot for training manipulation policies and rapid testing, with a strong focus on domain randomization and sim-to-real transfer.
• Design and train learning-based policies for bimanual manipulation in hospitality settings, and bring them from simulation to deployment on real hardware.
• Develop the 3D perception that drives manipulation, including multi-view depth and point-cloud processing, 6-DoF pose and grasp estimation, and scene representations built from RGB-D.
• Develop grasp strategies for the objects our robots encounter in service environments, including rigid and deformable items handled from imperfect real-world perception.
• Validate on real hardware through iterative cycles of refinement and redeployment, working with the perception and control engineers to integrate the manipulation stack.

Conoscenze

Python

PyTorch

ROS 2

Imitation learning

Teleoperation data collection

Point clouds

6-DoF pose estimation

Strumenti

Isaac Sim

MuJoCo

ManiSkill

Descrizione del lavoro

AICAM Mondovi Site (Cuneo) is looking for aVision Engineer - Learning-Based Robotic Manipulation Ha le carte in regola per avere successo? Le seguenti informazioni devono essere lette attentamente da tutti i candidati.

About AICAM

About AICAMAICAM () is Raicam's robotics division, focused on developing and industrializing service robots for real hospitality environments.

About the role

We are looking for a Vision Engineer - Learning-Based Robotic Manipulation (f/m/d) to help develop the learning-based manipulation capabilities of our robots, from perception to policy to execution on real hardware. You will work on the pipeline that turns multi-camera 3D perception into activity reasoning and robust bimanual manipulation, closing the loop by iteratively deploying on the real robot. You will join our robotics team and work closely with the perception, control, and hardware engineers who own the surrounding stack, focusing on the computer vision and robot learning core that enables our robots to carry out manipulation tasks in dynamic, unstructured hospitality environments such as restaurants, hotels, and hospitals.

Responsibilities

• Own the teleoperation and demonstration data pipeline: the collection setup, data quality and curation, and the path from recorded episodes to trained policies.
• Build high-fidelity simulation of the robot for training manipulation policies and rapid testing, with a strong focus on domain randomization and sim-to-real transfer.
• Design and train learning-based policies for bimanual manipulation in hospitality settings, and bring them from simulation to deployment on real hardware.
• Develop the 3D perception that drives manipulation, including multi-view depth and point-cloud processing, 6-DoF pose and grasp estimation, and scene representations built from RGB-D.
• Develop grasp strategies for the objects our robots encounter in service environments, including rigid and deformable items handled from imperfect real-world perception.
• Validate on real hardware through iterative cycles of refinement and redeployment, working with the perception and control engineers to integrate the manipulation stack.

Requirements

• Strong background in learning-based robotic manipulation, with hands-on experience on real hardware.
• Solid computer vision and 3D learning: point clouds, depth, multi-view geometry, and 6-DoF pose estimation or grasp regression.
• Hands-on experience with robotic grasping and/or imitation learning.
• Experience with teleoperation and data collection for robot learning.
• Proficiency in Python and PyTorch, with ROS 2 familiarity.
• Working experience with at least one robotics simulator (Isaac Sim / Lab, MuJoCo, ManiSkill).
• Proven ability to debug and tune robotic systems in real-world conditions, dealing with sensor noise, latency, and mechanical limitations.

Nice to have

• Experience in any of the following is a plus. We do not expect one person to cover all of them: Vision-Language-Action (VLA) models: fine-tuning or deploying policies such as π0 / π0.5, OpenVLA, SmolVLA, ACT, or Diffusion Policy.
• Reinforcement learning for manipulation, with sim-to-real transfer.
• GPU-accelerated simulation environments (Isaac Lab, ManiSkill, MuJoCo MJX).
• Advanced 3D scene representations for manipulation (Scene Graphs, NeRF, 3D Gaussian Splatting).
• Hands-on experience with robot embodiments such as mobile manipulators, dual-arm systems, or humanoids.

What we are looking for

Engineers who bridge theory and real-world execution, comfortable both reading a research paper and diagnosing why a policy fails on the real robot. A focus on robustness and generalization over ideal-case performance or cherry-picked demos. A strong ownership mindset on complex subsystems, with the initiative to push back when something is not working. Genuine interest in solving open perception and manipulation problems.

Compensation & Classification

Bargaining Agreement: Italian National Collective Bargaining Agreement (CCNL Metalmeccanico Industria). Classification Target: Expected job level B2, corresponding to a statutory base minimum of € 33058,74 gross per year. Total Rewards Package: Final financial offers will clearly exceed collective bargaining minimums to reflect individual expertise and prior achievement. The complete rewards proposal will include an individual merit allowance on top of the base minimum, a fixed production allowance, welfare benefits, performance bonuses, healthcare coverage (Metasalute), and pension options (Cometa).

We are an Equal Opportunity Employer. xdwybme Applications are welcome from all individuals regardless of gender, age, sexual orientation, ethnicity, background, or disability status.

Vision Engineer - Learning-Based Robotic Manipulation (f/m/d) Rome, Italy