Robot Learning Engineer
At a glance
Mid-level AI Research role at Dexmate. Singapore Office · full-time.
Pay not stated AI Research salary
aimlroles summary, based on the employer's posting.
What you'll do
- Train and assess learned policies for robot manipulation and whole-body control using imitation, reinforcement or VLA-style methods
- Run models on physical robots, tuning their behavior and linking policy decisions with controllers and hardware
- Move experiments between simulation and real robots, gathering data and investigating failures to improve results
- Build training infrastructure for GPU-cluster workloads, including scheduling, checkpoints, metrics and logs
- Create debugging and experiment-analysis tools, then coordinate with controls, embedded and hardware teams on deployment
What you bring
- A master's or doctorate in robotics, machine learning, computer science or a related area, or equivalent practical experience
- Practical experience training and deploying machine-learning models, preferably with PyTorch
- A background in robot learning, imitation or reinforcement learning, or learning-assisted control
- Experience moving models beyond offline testing into deployment, plus the ability to build and operate systems end to end
Who this fits
This role suits a mid-level engineer focused on policy learning, training infrastructure or real-time inference. You’ll work across research and physical robot deployment, collaborating with controls, embedded and hardware teams. The position is onsite in Singapore.
From the employer
About Dexmate
Dexmate is building the foundation for physical AI — combining a new generation of robots with a universal Physical AI OS, making robots as easy to build and deploy as software. Today, robotics is fragmented, slow, and closed: most builders are forced to reinvent the same stack again and again, and most ideas never make it past the prototype stage. We exist to change that. Our mission is to democratize robotics by lowering the barrier to entry, delivering a plug-and-play platform for developers, researchers, and enterprises, and cultivating an open ecosystem that accelerates the evolution of physical AI.
If you want to help shape the next layer of human capability — and believe the future of robotics should be built together, not in isolation — we'd love to build it with you.
The Role
We're developing end-to-end learned models for general-purpose robot manipulation and control, and you'll help build and shape the foundational systems that put capable robots into the real world. This is a broad role that can be tailored to your area of expertise — policy learning, training infrastructure, or real-time inference. You'll work across the full robot learning pipeline: ingesting and processing multimodal data, training large-scale models, deploying policies to physical hardware, and closing the loop between research and real-world performance.
Responsibilities
Robot Learning & Policy Development
Design, train, and evaluate learned policies for manipulation and whole-body control — imitation learning, reinforcement learning, and VLA-style architectures
Deploy and validate models on physical robots, tuning for real-world performance and closing the loop between policy, controller, and hardware
Iterate rapidly between simulation and real robots — design experiments, collect data, debug failure modes, and drive measurable improvements
Own infrastructure for model training: job scheduling, checkpointing, metrics, and logging; scale distributed training across GPU clusters
Build research tooling for debugging, visualization, and experiment analysis
Work closely with controls, embedded, and hardware teams to translate research needs into reliable, deployable systems
Minimum Qualifications
Master's or PhD in Robotics, ML, CS, or related field, or equivalent practical experience
Hands-on experience training and deploying ML models, ideally in PyTorch
Background in robot learning, imitation/reinforcement learning, or learning-augmented control
Experience taking models from research to deployment — not just offline benchmarks
Ownership mindset: you design, build, operate, and iterate on systems end-to-end
Preferred Qualifications
Experience with VLAs, video generation architectures, or robot foundation models
Experience deploying learned policies on physical robots — manipulation, legged, or humanoid systems
Experience with serialization formats for high-performance systems (Protobuf, MCAP)
Track record of publications at top-tier venues (ICRA, IROS, CoRL, RSS, NeurIPS, ICML) is a strong plus