New opportunity

Research Associate (Embodied AI)

Nanyang Technological University · Singapore

About this role

Nanyang Technological University’s School of Electrical and Electronic Engineering seeks a Research Associate to support research and engineering in perception and embodied intelligence. The project focuses on dexterous hand robotic manipulation for household service and manufacturing.

Responsibilities

Develop robotic foundation models for general dexterous hand manipulation using vision-language-action models.
Develop a manipulation-learning framework using human instructional videos to reduce the data needed for policy learning.
Build a dexterous hand manipulation system capable of diverse household tasks and deployment in manufacturing scenarios.

Requirements

Master’s degree in Electrical and Electronic Engineering, Computer Science, or a related field, plus 1 year of research experience in a related area.
Expertise in robotic manipulation systems and knowledge of robotic foundation models.
Basic proficiency with ROS and Python, and strong written and verbal communication skills for explaining complex concepts clearly and accurately.
A publication track record is an advantage, not a requirement.

The full-time position is located at NTU Main Campus, Singapore. Only shortlisted candidates will be notified.

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

Skills for this role

Robotic manipulationRobotic foundation modelsVision-language-action modelsManipulation learningPolicy learningROSPythonPerceptionEmbodied intelligenceResearch communication

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

Vision-language-action models for dexterous manipulation
The role calls for developing robotic foundation models for general dexterous hand manipulation based on vision-language-action models. Prepare to explain how perception, instructions, and robot actions connect in a manipulation approach.
Data-efficient learning from instructional videos
The posting specifies a learning framework that uses human instructional videos to reduce the data needed for policy learning. Prepare to discuss how video-derived information could support learning and how to evaluate data efficiency.
A system for household and manufacturing tasks
The project aims to build a dexterous hand system for diverse household tasks and manufacturing deployment scenarios. Prepare to discuss how model development would translate into a working manipulation system across those settings.
Clear research and engineering communication
The role supports both scientific research and engineering practice, and requires clear written and verbal explanations of complex concepts. Prepare a concise account of a related research approach, its evidence, and its limitations.

Likely challenges to think through

Turning human video into robot learning signals
A likely difficulty is that instructional videos show human behavior, not directly executable dexterous-hand actions; extracting useful signals must be balanced against the goal of reducing robot policy-learning data. Prepare to explain a possible representation and how you would test whether it helps.
Generalizing across tasks and settings
A likely trade-off is between a broadly capable model and reliable performance on individual tasks in both household and manufacturing scenarios. Prepare to describe how you would assess performance across varied tasks without assuming one setting predicts the other.
Connecting research models to a working system
The posting combines foundation-model research with development of a dexterous manipulation system. A likely challenge is separating model limitations from perception, control, or integration failures; prepare to describe staged tests that would help identify the cause.

Practice your answers

Technical practice

How would you approach dexterous hand manipulation for a range of tasks?

What to coverThe role involves building a dexterous hand system for diverse household tasks and manufacturing scenarios. Preparation suggestion: explain how you would coordinate perception, hand actions and feedback, including trade-offs between precise manipulation and robustness.

What perception information would a dexterous manipulation policy need?

What to coverThe posting places the work in perception and embodied intelligence. Preparation suggestion: discuss how visual observations could inform object state and hand actions, and how you would handle ambiguous or changing observations.

How would you formulate a vision-language-action model for hand manipulation?

What to coverDeveloping robotic foundation models based on vision-language-action models is a stated responsibility. Preparation suggestion: describe the roles of visual input, task language and action output, then explain a plausible training objective and its limitations.

How would you test whether a manipulation foundation model generalizes?

What to coverThe posting calls for general dexterous hand manipulation rather than a single task. Preparation suggestion: compare performance on familiar and held-out tasks or objects, and explain how you would distinguish genuine transfer from memorization.

How could human instructional videos support manipulation learning?

What to coverThe role explicitly involves learning from human instructional videos. Preparation suggestion: explain what task information videos can provide, what robot actions they do not directly specify, and how you would address that gap.

How would you measure whether a learning framework reduces policy-training data demand?

What to coverReducing the data needed for policy learning is an explicit project aim. Preparation suggestion: compare policies at several data budgets under consistent evaluation conditions, and discuss the trade-off between data efficiency and task performance.

How would you evaluate a dexterous hand system across household and manufacturing scenarios?

What to coverBoth diverse household tasks and manufacturing deployment scenarios appear in the responsibilities. Preparation suggestion: propose task-success and reliability measures, and explain why evaluation conditions may need to differ across the two settings.

How would you organize a ROS and Python implementation for a manipulation experiment?

What to coverThe posting asks for basic proficiency with ROS and Python. Preparation suggestion: outline how you would connect observations, policy outputs and robot commands, while making experiments reproducible and failures observable.

How would you diagnose a manipulation policy that works in testing but fails on a new task?

What to coverThe role combines policy learning with a system intended to complete diverse tasks. Preparation suggestion: separate possible perception, model-generalization and robot-execution causes, then describe tests that would narrow down the failure.

Behavioural practice

Tell me about a research question you turned into a testable experiment.

What to coverThe role supports both scientific research and engineering practice. Preparation suggestion: use a real situation, explain the hypothesis and actions you took, and report the result, including what you changed after seeing the evidence.

Describe a time you debugged an integrated technical system.

What to coverBuilding a working dexterous manipulation system is part of the advertised work. Preparation suggestion: give a real example with the failure context, your diagnostic actions and the observed result; distinguish your contribution from others'.

Give an example of explaining a complex technical finding clearly.

What to coverThe posting explicitly requires clear, accurate written and verbal communication. Preparation suggestion: describe the audience and situation, how you adapted your explanation, and the result or feedback that showed whether it worked.

Tell me about a time you balanced a research idea with an engineering constraint.

What to coverThe position spans research and engineering, including a goal of reducing data demand for learning. Preparation suggestion: choose a real situation, explain the competing considerations and your actions, then state the measured result or remaining limitation.

Sharpen your knowledge

Topics to prepare

Dexterous-hand manipulation
The posting requires expertise in robotic manipulation systems. Revise how a hand uses perception and feedback to grasp, reposition and release an object; sketch where a manipulation attempt could fail.
Robotic foundation models and vision-language-action models
The role involves developing foundation models for general dexterous-hand manipulation based on vision-language-action models. Trace how a visual observation and language instruction could lead to a hand action, and identify what would need to generalize across tasks.
Learning from human instructional videos
The project aims to use instructional videos to reduce the data needed for policy learning. Compare what a video visibly demonstrates with the information a robot policy would need; explain one possible data-efficiency approach as a preparation suggestion, not a specified method.
Household and manufacturing task transfer
The planned system targets diverse household tasks and manufacturing deployment scenarios. Compare two hypothetical tasks and list which manipulation demands might transfer and which would require separate evaluation. The posting does not specify particular tasks.

Skills to revise

Python and ROS basics
The posting asks for basic proficiency in both. Practise a small Python-based ROS message exchange representing a hand command and observed state, then explain the message flow; no ROS version or interface is specified.
Instructional-video analysis
To prepare for the stated video-based learning work, take a short instructional clip and make a timestamped table of visible goals, object interactions and ambiguities. Treat the table format as practice, not a required project method.
Manipulation-system evaluation
As a suggested exercise for the system-development responsibility, draft a test sheet for two hypothetical hand-manipulation tasks: define observable success, record failures and note whether performance changes between settings. These are not specified employer benchmarks.
Clear research communication
The posting requires clear written and verbal explanations of complex concepts. Write a brief explanation of how video-based learning might reduce policy-training data, then practise a spoken version that separates the proposed mechanism from unverified results.

Your preparation checklist

Map your evidence to the stated requirements
Prepare a one-page summary of your relevant degree, research experience, robotic manipulation work, foundation-model knowledge, and basic ROS and Python proficiency. Distinguish direct experience from topics you have only studied.
Prepare a manipulation project walkthrough
If you have a relevant project, bring a concise account of the task, hand and sensing setup, your contribution, results, and failures. Be ready to explain what would change for household tasks or manufacturing deployment.
Sketch a vision-language-action approach
As a proposed interview exercise, draw a pipeline from visual observations and task instructions to dexterous-hand actions. Explain how you would evaluate generalization and where the approach might fail; the posting specifies the research direction, not an architecture.
Design a video-to-policy learning experiment
Outline how human instructional videos could inform manipulation learning, then propose a comparison that measures whether the approach reduces policy-training data demand. State assumptions about video quality and available robot data.
Prepare an implementation and evaluation plan
Draft a small ROS-and-Python system diagram and a test plan for diverse household tasks and manufacturing scenarios. Mark hardware, datasets, and deployment constraints as questions rather than assuming NTU’s setup.
Practice explaining your research clearly
Prepare a short spoken explanation and a written summary of one complex result. Bring publications if you have them; the posting calls a publication track record an advantage, not a requirement.
Write questions for the research team
Ask which dexterous-hand platform and data are available, how success and data efficiency will be measured, and what distinguishes the intended household and manufacturing scenarios.

Availability last checked: 4 Nov 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…