Lead AI Research Scientist (Robotics)

Asiacruit BPO, Inc.

18h•Hybrid

About The Position

Asiacruit is seeking a Lead Physical AI / Vision-Language-Action (VLA) Research Scientist to lead development and adaptation of foundation models that connect visual perception, language understanding, and robotic actions. The role guides research and implementation to help robots interpret visual information, understand natural-language instructions, and generate actions for navigation and manipulation. It combines multimodal AI, foundation model adaptation, robotic learning, and applied computer vision. The scientist will lead model adaptation, research planning, technical evaluation, experimentation, performance improvement, and guidance of researchers and engineers through model training, evaluation, and deployment. The role collaborates with robotics controls and embedded systems teams to translate model outputs into executable robot behaviors. The position can be remote through a US-based research team, subject to the employer’s approved work locations, or hybrid in New Delhi, India.

Requirements

Master's or Ph.D. in Computer Science, Artificial Intelligence, Robotics, Machine Learning, or a related field.
At least 4 years of relevant experience in multimodal AI, foundation models, vision-language models, imitation learning, or robotic learning.
Strong experience developing, training, or fine-tuning large multimodal models, Vision Transformers, or related deep learning architectures.
Strong proficiency in Python and PyTorch or JAX.
Hands-on experience with vision-language models or Vision-Language-Action (VLA) architectures.
Familiarity with open-weight VLA models such as OpenVLA, Octo, or π0.
Experience with Hugging Face tools and model training or fine-tuning workflows.
Practical knowledge of parameter-efficient fine-tuning techniques, including LoRA or related methods.
Understanding of multimodal representation learning, visual perception, and language-conditioned action generation.
Familiarity with robotic imitation learning, policy learning, or reinforcement learning.
Ability to lead technical research, evaluate experimental results, and translate findings into practical AI capabilities.

Nice To Haves

Understanding of cross-embodiment robotic learning and datasets such as Open X-Embodiment.

Responsibilities

Lead the adaptation and development of Vision-Language-Action models for robotic applications.
Fine-tune open-weight VLA models such as OpenVLA, Octo, and π0 for navigation and manipulation tasks.
Develop multimodal representations connecting camera observations, natural-language instructions, spatial reasoning, and robot actions.
Design action-tokenization and spatial waypoint generation approaches for robotic behavior.
Develop parameter-efficient fine-tuning pipelines using LoRA and other PEFT methods.
Work with vision-language models, Vision Transformers (ViT), and large multimodal models.
Integrate visual perception and AI capabilities into robotic decision-making systems.
Explore cross-embodiment learning and datasets such as Open X-Embodiment to improve generalization across robotic platforms.
Collaborate with robotics controls and embedded systems teams to translate model outputs into executable robot behaviors.
Lead research planning, technical evaluations, experiments, and model performance improvements.
Guide researchers and engineers in model adaptation, training, evaluation, and deployment.