Beewise builds autonomous robotic beekeeping systems that operate in commercial fields. We are looking for a Physical AI Engineer to turn simulation and teleoperation data into vision-language-action (VLA) policies that can handle real beehives.

This is a hands-on role with research depth, working in a small, innovative team to develop, evaluate, and deploy robot-learning solutions in real-world conditions.

Location: On site at Beewise, Beit HaEmek, near Nahariya, northern Israel.

Responsibilities

- Fine-tune VLA and world models for bimanual, contact-rich manipulation, and optimize them for embedded edge compute.

- Build physical simulation and the digital twin into scalable data sources and close the sim-to-real gap.

- Follow advances in Physical AI and robot learning, and run focused experiments to assess what is worth adopting.

- Shape teleoperation data collection and build tools to curate and combine real and simulated episodes.

- Evaluate policies in simulation and on physical robots, and trace failures to data, models, simulation, or hardware.

- Develop and implement autonomous performance evaluation methods to assess system operation, detect abnormal behavior, identify failure modes, and trigger safe shutdown and escalation when necessary.

• An advanced degree in a relevant science/engineering field.

• Strong foundations and hands-on experience in training and evaluating deep learning models.

• Experience in Reinforcement Learning, including its application to robot learning and manipulation.

• Curiosity and an open mind, with the ability to learn and explore emerging Physical AI approaches.

• An innovative, research-driven approach: form independent views from papers and first principles, then test them.

• Ability to scope open-ended problems and deliver working solutions independently.

• Clear and honest communication of experimental results, limitations, and lessons learned.

• Strong collaboration and communication skills across mechanical, embedded, and product teams.

Experience training and deploying manipulation policies on a physical robot.

Nice to have

• Hands-on experience with imitation learning, VLA models, world models, and Reinforcement Learning.
• Experience with edge deployment, including distillation, quantization, or inference optimization.
• Familiarity with robotics middleware
• Familiarity with robot-learning data formats and tooling.
• Publications or open-source contributions in robotics or physical AI topics.