Lucifer Robotics

is a Bengaluru-based research lab solving dexterous manipulation. Our mission is to bring machines the power of touch on the path to physical general intelligence, bringing about a future of abundance.

Member of Technical Staff — Pre-Training

As a founding hire, you will own the architecture and training recipe for our visuo-tactile foundation model: the core representation that lets a robot understand the physical world through both sight and touch. This is the central modeling bet of the lab, and you will set its direction.

What You'll Do
- Own the model architecture end to end: how visual and tactile signals are tokenized, fused or kept separate, and learned jointly.
- Design the training recipe across a staged pipeline — large-scale pretraining on passive observational data, a mid-training stage on our proprietary contact-rich data, and downstream adaptation.
- Set evaluation:



define what "this model understands contact" actually means and build the evals for it.
- Set the scaling roadmap as data and compute grow.

You Might Be a Fit If You
- Have trained large multimodal or sequence models from scratch and owned the recipe — the data mix, curricula, ablations, debugging silent failures.
- Have robust instincts for multimodal fusion and have shipped a model where a second modality measurably earned its place.
- Are comfortable being the person who decides the architecture when the literature gives no clean answer.
- Bonus: tactile, force, or other non-visual sensing; world models; self-supervised representation learning.
- We seek a high-ownership individual up to the task of defining this vision. This is a founding research role and will be compensated accordingly with equity.

📌 Robotics Pre-Training Engineer (Bengaluru)
🏢 Lucifer Robotics
📍 Bengaluru