Robot Learning (VLA) - Internship

About the internship:Selected intern's day-to-day responsibilities include:

1. Train, fine-tune, and evaluate Vision-Language-Action (VLA) models on collected datasets using PyTorch and the Hugging Face ecosystem

2. Apply parameter-efficient fine-tuning (LoRA, QLoRA, PEFT) and quantization where appropriate.

3. Analyze model failures and iteratively improve datasets, augmentation strategies, and training configurations.

4. Build and configure Gazebo and NVIDIA Isaac Sim / Isaac Lab environments to evaluate policies before hardware deployment.

5. Deploy and optimize models on GPU and edge platforms, including CUDA, TensorRT, ONNX, and Jetson, for real-time control.

6. Integrate models with ROS 2, sensors, actuators, and robot control pipelines. Who can apply:Only those candidates can apply who:are available for full time (in-office) internshipcan start the internship between 29th Sep'26 and 3rd Nov'26are available for duration of 6 monthsare from Noida onlyhave relevant skills and interestsare Computer Science Engineering studentsStipend:INR₹ 20,000 - 30,000 /monthDeadline:[hidden]:59:59Other perks:Certificate, Letter of recommendation, Informal dress code, Free snacks & beveragesSkills required:OpenCV, Python, Linux, Git, Machine Learning, Model View Controller(MVC), Artificial intelligence, Gazebo, PyTorch, Model Evaluation, NumPy, Hugging Face and Model fine-tuningOther Requirements:1. Final-year engineering student with hands-on experience building and deploying real end-to end systems, beyond API-based projects.2.VLA & robotics foundation models: OpenVLA, π0, ACT, SmolVLA, Diffusion Policy, VLMs, and multimodal models.3. Model fine-tuning: LoRA, QLoRA, and PEFT for open-source models such as Llama, Qwen, Gemma, Mistral, or similar.4. Inference optimization: Quantization, ONNX, TensorRT, and CUDA.5. Simulation: Gazebo, RViz, NVIDIA Isaac Sim, or Isaac Lab.About Company:Variety Innovation Venture Private Limited is a Delhi based research & development and industrial automation company. We bring most dimensions of innovation together to work in new ways, enter new categories or channels and produce new products and services so as to bring new fronts. We offer a complete range of innovation services in various sectors like industrial automation, consultation, R&D, etc. With a team of experienced engineers with a specialization in mechatronics engineering, software engineering, controls, automation, and robotics, we are able to provide efficient systems integration solutions for complex engineering challenges. We had worked in various sectors of industry with all kinds of companies and organizations, on a remarkable range of projects. The diversity of our knowledge and skills gives a clear advantage to our clients, as we combine the learning and experience gained, to help build new breakthrough solutions.

Responsibilities

Selected intern's day-to-day responsibilities include:

1. Train, fine-tune, and evaluate Vision-Language-Action (VLA) models on collected datasets using PyTorch and the Hugging Face ecosystem

2. Apply parameter-efficient fine-tuning (LoRA, QLoRA, PEFT) and quantization where appropriate.

3. Analyze model failures and iteratively improve datasets, augmentation strategies, and training configurations.

4. Build and configure Gazebo and NVIDIA Isaac Sim / Isaac Lab environments to evaluate policies before hardware deployment.

5. Deploy and optimize models on GPU and edge platforms, including CUDA, TensorRT, ONNX, and Jetson, for real-time control.

6. Integrate models with ROS 2, sensors, actuators, and robot control pipelines.