Position Summary

We are seeking a talented Robot Vision-Language-Action (VLA) Engineer to design, implement, and deploy end-to-end VLA systems that bridge multimodal foundation models with embodied robotic agents. You will work at the intersection of large language models, computer vision, reinforcement learning, and low-level control to create robots that can follow high-level human instructions like "organize the kitchen by putting away dishes and wiping counters" with minimal supervision.

Key Responsibilities

• Model Architecture & Integration

◦ Design and train VLA models that fuse vision encoders (e.g., CLIP, DINOv2, SigLIP), language backbones (e.g., LLaMA-3, Qwen), and action heads (e.g., diffusion policies, ACT, RT-2 style tokenizers).

◦ Implement efficient inference pipelines for real-time operation on edge hardware (e.g., NVIDIA Jetson, custom ASIC/FPGA accelerators).

• Data Engine & Synthetic Pipelines

◦ Build large-scale datasets combining real robot trajectories, teleoperated demos, and simulated data (Isaac Sim, Mujoco, PyBullet).

◦ Develop data augmentation strategies: video infilling, language rephrasing, 3D scene randomization, and cross-embodiment transfer.

• Training & Scaling

◦ Scale pretraining on internet-scale video + text + action triples (e.g., Ego4D, Something-Something, BridgeData).

◦ Fine-tune with reinforcement learning (RLHF, PPO, DPO) and imitation learning (BC, IQL) using reward models for affordance, safety, and task success.

• Robot Deployment & Evaluation

◦ Deploy VLA policies on physical platforms (manipulators, mobile manipulators, humanoids).

◦ Create standardized benchmarks: zero-shot generalization, long-horizon task completion, sim-to-real gap, language ambiguity resolution.

• Safety & Reliability

◦ Implement guardrails: force/torque monitoring, collision prediction, verbal clarification loops, and reversible action primitives.

◦ Conduct failure mode analysis and iterative red-teaming.

Required Qualifications

• MS/PhD in Computer Science, Robotics, or related field (or equivalent experience).

• Experience in one or more of:

◦ Training/inference of vision-language models (CLIP, FLAVA, LLaVA, etc.).

◦ Robot learning from demonstration (ACT, Diffusion Policy, RT-1/2).

◦ Reinforcement learning for manipulation (DrQ-v2, R3M, VIP).

• Strong Python; proficient in PyTorch/JAX.

• Hands-on experience with ROS2, Docker, and real robot hardware.

Preferred Skills

• Published work at CoRL, RSS, ICRA, NeurIPS, or CVPR on VLA, robot foundation models, or multimodal learning.

• Experience with 3D vision (NeRF, Gaussian splatting, point cloud processing).

• Low-level control (impedance/admittance control, MPC, dynamics modeling).

• Distributed training on GPU clusters; familiarity with SLURM, Ray, or Kubernetes.

What We Offer

• Competitive salary + equity

• Health, dental, vision

• Relocation support

• Access to a state-of-the-art robot lab (multiple arms, mobile bases, high-fidelity sim)

• Collaboration with world-class researchers in VLMs and robotics

Location

Kowloon, Hong Kong

SZ, China

How to Apply

Send your CV, GitHub/portfolio, and a 1-page research statement describing your vision for the next generation of VLA systems to careers@r2c2.ai with subject "Robot VLA Engineer – [Your Name]".

We are an equal-opportunity employer and value diversity at our company.