New opportunity

Robotics Engineer - Singapore: Foundation Model Training

General Robotics · Singapore

About this role

General Robotics is seeking Research Engineers to develop foundation models for their GRID platform. The role involves working across model architecture, large-scale training, reinforcement learning, simulation, and evaluation to build robotic behaviors that generalize across tasks, environments, and embodiments. Responsibilities include formulating research hypotheses, designing experiments, developing and scaling foundation models for perception and action, leveraging simulation for pretraining and evaluation, and advancing sim-to-real transfer. Candidates will implement and communicate research as production-quality work. Minimum qualifications include a Bachelor's degree in CS, Robotics, or a related field, research or applied experience in ML/Robotics/Computer Vision, strong Python and PyTorch skills, experience with robotics simulation, and familiarity with inverse kinematics, dynamics, and manipulation. Demonstrated research ability through publications or industry work is required. Desired qualifications include an MS or PhD, first-author publications at top-tier conferences (NeurIPS, ICML, ICLR, CVPR, CoRL, RSS, ICRA), and hands-on experience with VLA models, world models, diffusion/flow-matching policies, transformers, and large-scale pretraining. Experience with specific simulation platforms like NVIDIA Isaac Sim, MuJoCo, or ManiSkill is highly valued. The Singapore team functions as an extended research facility collaborating closely with the US headquarters.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

PythonPyTorchMachine LearningRoboticsComputer VisionReinforcement LearningSimulationInverse KinematicsDynamicsRobotic ManipulationVision-Language-Action modelsWorld modelsDiffusion modelsFlow-matching policiesTransformersImitation learningNVIDIA Isaac SimIsaac LabMuJoCoManiSkill

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Foundation Model Development and Scaling
The core responsibility is to design, train, and scale foundation models for robotic perception and action. You must demonstrate the ability to handle large-scale pretraining, data-mixture design, and post-training processes, specifically using PyTorch to build models that generalize across diverse robotic embodiments and tasks.
Simulation-to-Real Transfer and Evaluation
You must leverage simulation platforms like NVIDIA Isaac Sim, MuJoCo, or ManiSkill to generate data and train policies. A key priority is bridging the gap between simulated environments and real-world deployment, ensuring that models remain robust and reliable when applied to physical robotic systems.
Research-to-Production Implementation
The role requires translating high-level research hypotheses into production-quality, reusable code. You must be able to communicate research findings clearly and ensure that experimental models are sufficiently robust for deployment, moving beyond theoretical prototypes to scalable engineering solutions.

Likely challenges to think through

Bridging Cross-Continental Research Collaboration
As the Singapore team functions as an extended facility for the US headquarters, you will face the challenge of maintaining tight alignment on research goals across time zones. Prepare by highlighting your experience in distributed team environments and your ability to document research progress clearly to ensure seamless integration with the US-based team.
Generalization Across Heterogeneous Embodiments
Building models that generalize across different robot types (e.g., arms vs. humanoids) is inherently difficult due to varying dynamics and kinematics. Prepare by discussing your approach to designing model architectures that are embodiment-agnostic and how you handle diverse data distributions during the training phase.

Practice your answers

Technical practice

Sim-to-Real Transfer Strategies

What to coverDiscuss techniques to bridge the reality gap when deploying models trained in simulation to physical hardware. Focus on domain randomization, system identification, or architectural choices that improve robustness to real-world sensor noise and dynamics.

Vision-Language-Action (VLA) Model Architecture

What to coverExplain the design considerations for VLA models in robotics, specifically how to align visual perception with language instructions and motor control outputs. Address the trade-offs between end-to-end training and modular approaches.

Scaling Simulation for Data Generation

What to coverDescribe how you would scale a simulation platform like NVIDIA Isaac Sim or MuJoCo to generate diverse datasets for reinforcement learning. Discuss strategies for managing heterogeneous scenes and tasks to ensure model generalization.

Diffusion and Flow-Matching Policies

What to coverCompare diffusion models and flow-matching policies for robot control. Explain why these generative approaches are increasingly favored over traditional behavioral cloning for complex, multi-modal manipulation tasks.

Reinforcement Learning for Manipulation

What to coverDetail your experience with RL in robotic manipulation. Focus on reward function design, handling sparse rewards, and the challenges of training policies that must generalize across different robotic embodiments.

Transformer Applications in Robotics

What to coverDiscuss the role of Transformer architectures in processing temporal sequences for robotic action. Explain how attention mechanisms help in understanding spatial relationships or long-horizon task planning.

Inverse Kinematics and Dynamics Integration

What to coverExplain how you integrate classical robotics concepts like inverse kinematics and dynamics with modern learning-based approaches. Suggest how these constraints can be used to regularize or guide model training.

Large-Scale Pretraining and Data Mixtures

What to coverDescribe your approach to designing data mixtures for pretraining foundation models. Address how you balance different data sources to ensure the model learns both generalizable features and task-specific precision.

Behavioural practice

Translating Research to Production

What to coverDescribe a time you took a research hypothesis through to a production-quality implementation. Focus on how you handled the transition from experimental code to a robust, reusable system.

Cross-Site Collaboration

What to coverThe Singapore team works closely with the US headquarters. Provide an example of how you have successfully collaborated with a remote team across different time zones to achieve a shared research or engineering goal.

Handling Failed Experiments

What to coverResearch involves high uncertainty. Share an example of a research experiment that did not yield the expected results and explain how you analyzed the failure to pivot your approach.

Prioritizing Research vs. Deployment Needs

What to coverExplain how you balance the need for long-term research innovation with the immediate requirements of real-world deployment. Provide an example of how you made a trade-off decision in a project.

Sharpen your knowledge

Topics to prepare

Sim-to-Real Transfer Methodologies
The role emphasizes improving reliability under real-world constraints. Review literature on domain randomization, domain adaptation, and system identification to bridge the gap between simulation and physical deployment. Exercise: Draft a summary of how you would mitigate the reality gap when deploying a policy trained in NVIDIA Isaac Sim to a physical robotic arm.
Foundation Model Architectures for Robotics
Focus on the intersection of Vision-Language-Action (VLA) models, world models, and diffusion-based policies. Understand how transformers are applied to multimodal robotic control. Exercise: Compare the architectural trade-offs between diffusion policies and autoregressive transformer models for continuous action space control in manipulation tasks.
Reinforcement Learning and Imitation Learning
The position requires scaling RL and leveraging datasets for pretraining. Review state-of-the-art techniques in offline RL and behavior cloning. Exercise: Design an experiment to evaluate the sample efficiency of an imitation learning agent versus an RL agent in a simulated manipulation environment.

Skills to revise

PyTorch for Large-Scale Model Training
The role involves developing and scaling foundation models. Practice implementing custom training loops, distributed data parallel (DDP) training, and mixed-precision techniques. Exercise: Create a PyTorch script that demonstrates a distributed training setup for a transformer-based policy model.
Robotics Simulation Platforms
Proficiency in platforms like NVIDIA Isaac Sim, MuJoCo, or ManiSkill is highly valued for data generation. Exercise: Build a simple manipulation task in MuJoCo or Isaac Sim that generates synthetic trajectory data for training a policy.
Inverse Kinematics and Dynamics
Fundamental understanding of robot control is required. Exercise: Implement a basic inverse kinematics solver for a 6-DOF robotic arm and validate its accuracy against a forward kinematics model in Python.

Your preparation checklist

Prepare a technical portfolio of research-to-production projects
Curate examples of your work that demonstrate taking a research hypothesis through to a robust, production-quality implementation. Focus on projects involving PyTorch, robotics simulation, or foundation models, highlighting your ability to write clean, reusable code.
Review your experience with robotics simulation platforms
Be prepared to discuss your hands-on experience with platforms like NVIDIA Isaac Sim, MuJoCo, or ManiSkill. Specifically, explain how you have used these tools for data generation, reinforcement learning, or sim-to-real transfer.
Synthesize your expertise in foundation models for robotics
Prepare to articulate your technical depth in areas such as Vision-Language-Action (VLA) models, world models, or diffusion/flow-matching policies. Be ready to explain the trade-offs you have encountered when scaling these models for robotic perception and action.
Formulate questions about the GRID platform and collaboration
Prepare thoughtful questions regarding the technical challenges of the GRID platform, the specific research focus of the Singapore facility, and how the team maintains alignment and collaboration with the US headquarters.

Availability last checked: 18 Oct 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in Singapore. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…