As an R&D Computer Vision intern, you will be part of the R&D Team to improve Inbolt’s tracking algorithms (accuracy, reliability, frame rate, etc.).
Responsibilities
- Explore various state-of-the-art computer vision techniques (Machine Learning, Deep Learning, Odometry, etc.) and assess their relevance to our industrial constraints (real-time, robustness, embedded compute)
- Design, develop and implement state-of-the-art positioning algorithms, from prototype to production‑grade code
- Benchmark and validate your algorithms against real robotic setups and industrial use cases, iterating based on measured accuracy and latency
- Collaborate closely with the R&D and engineering teams to integrate your work into Inbolt Studio
- Stay on the bleeding edge of research and participate actively in the community (scientific papers, patents)
Benefits
- Prime office location
- Team‑building events & off‑sites
- Bring-your‑dog‑to‑work policy
- Meal — lunch expenses
- Gymlib — subscription covered
- Health coverage
Inbolt values end‑to‑end ownership, high rigor, and a hands‑on curiosity. We move fast while staying precise, with a proactive, autonomous, results‑driven mindset and strong collaboration with engineering teams.
Preferred experience
- Currently in your final year (Bac+5) at an engineering school or university Master’s program, looking for a 5‑6 month end‑of‑studies internship (stage de fin d’études)
- Genuine curiosity for robotics and 3D vision, comfortable navigating ambiguity and open‑ended research questions
- Python (torch, TensorFlow, or similar is appreciated)
- Deep Learning and Computer Vision knowledge
- Capacity to handle a good level of code quality and stability
- Fluent or at least working‑level French and/or English
What we love
- Powerful team spirit and enjoy working closely with a small, fast‑moving team
- Strong working motivation - we have a sustained work pace
- Proactive and autonomous - “Get it done” attitude with the will to overreach objectives
- Rigorous about testing and validating your work before calling it done