Robotics Data Pipeline Intern – Multimodal Data
About Us At Persona, we're building the next generation of humanoid robots, and that requires an unprecedented volume of high-quality, multimodal data. We're moving beyond basic teleoperation to leverage massive datasets of in-the-wild egocentric video combined with dense sensor streams (IMU, haptics, kinematics, and high-fidelity force profiles). We're looking for a curious, technically sharp intern to roll up their sleeves and help us turn raw, unstructured multimodal data into high-fidelity training assets for our robots.
The Role As a Data Pipeline Intern, you'll work directly alongside our data and robotics engineering teams to support the infrastructure that feeds our foundation models. You'll get hands-on experience with real multimodal data challenges, from sensor stream processing and video pipeline optimization to force analysis and kinematic retargeting. This is not a "fetch coffee and shadow engineers" internship. You'll own real work and ship real code.
What You'll Work On
- Rebuilding and extending pipelines that ingest and synchronously process egocentric video alongside rich sensor streams (IMU, force-torque, tactile, proprioception)
- Owning post-processing algorithms for force analysis and hidden state inference, including contact force estimation, occlusion handling, and inverse kinematics gap-filling
- Bridging kinematic retargeting work that translates human hand tracking into humanoid end-effector coordinates
- Optimizing and testing data augmentation strategies (spatial, temporal, synthetic viewpoints, sensor noise injection)
- Tying together work across our Hardware Teleoperation Team to help align human-robot play-data across modalities
What We're Looking For
- Currently pursuing a B.S., M.S., or Ph.D. in Computer Science, Data Engineering, Machine Learning, Robotics, or a related field
- Solid Python skills and exposure to PyTorch, particularly around data loading or multimodal datasets
- Coursework or project experience with computer vision, time-series data, or sensor processing
- Familiarity with video processing tools (OpenCV, FFmpeg) or pose estimation frameworks (MediaPipe) is a plus
- Awareness of imitation learning, VLA architectures, or human-to-robot transfer concepts is a plus, but genuine curiosity counts for a lot here
Bonus Points
- Experience with NVIDIA's robotics stack (Isaac, Cosmos, GR00T)
- Exposure to distributed computing (Ray, Spark) or simulation environments (Omniverse, MuJoCo)
- Any project work involving synthetic data generation or tactile/spatial data representations