Physical AI is the hardest hiring problem in technology right now, and it is not close.
The pool is small, almost nobody in it is applying for jobs, and the people evaluating candidates frequently cannot tell from a CV whether someone has shipped a robot that works or only built one that demoed. This is what these searches actually look like.
What physical AI hiring covers
"Physical AI" and "embodied AI" describe AI that acts in the world — robots, autonomous vehicles, drones, warehouse automation, industrial machines — rather than AI that generates text or images. The hiring problem is different because the required skills sit at an intersection very few people occupy:
- Perception — computer vision, sensor fusion, LiDAR, depth, calibration
- SLAM and localisation — knowing where the machine is, reliably, when sensors lie
- Controls and motion planning — making it move without breaking anything
- Embedded and firmware — real-time systems, RTOS, CAN bus, board bring-up
- Robotics ML — learned policies that have to survive physical reality
A strong ML engineer who has never handled sensor noise or a real-time control loop is not a substitute. That substitution is the most common and most expensive hiring mistake in the field.
The pool is genuinely small
Most "talent shortage" claims are really reach problems in disguise. Robotics is one of the few places where the shortage is literal.
For a specific profile — someone who has shipped a production perception stack, or owned calibration end to end, or taken an AMR fleet from prototype to deployment — the credible pool in any single market is measured in hundreds. Not thousands.
And they are almost all employed, at companies building things they find interesting, which is the hardest possible candidate to move. They are not browsing job boards. The proportion of our robotics placements who were not applying anywhere when we found them is higher than our 72% overall average.
Where they actually are
Not on job portals, mostly. Five channels, in rough order of yield:
Their published work. GitHub repositories, ROS and ROS2 contributions, SLAM and vision papers, conference talks, university lab affiliations. In robotics more than anywhere else, what someone has built is public and legible, and it tells you more than a CV.
Domain communities. Robotics, perception and embedded groups — mostly closed Slack and WhatsApp channels and specialist forums. Not places you can advertise into; places you have to already be in.
Named target companies. The robotics companies, AI labs and automotive and drone platforms in your geography are a finite, knowable list. A real robotics search starts with that list, person by person.
Research labs and PhD pipelines. Particularly for perception and learned control, where the strongest people often come directly out of a lab.
Adjacent industries. Automotive, aerospace, industrial automation and defence hold serious controls and embedded talent that robotics teams routinely overlook.
What it costs
India, 2026, total compensation. Deep-tech and venture-backed companies sit at the top of each band.
| Role | Mid | Senior | Lead / Principal |
|---|---|---|---|
| Perception / computer vision | ₹22–40 L | ₹40–70 L | ₹70 L–1.4 Cr |
| SLAM / localisation | ₹24–42 L | ₹42–75 L | ₹75 L–1.5 Cr |
| Controls / motion planning | ₹20–36 L | ₹36–62 L | ₹62 L–1.2 Cr |
| Embedded / firmware | ₹16–30 L | ₹30–52 L | ₹52 L–95 L |
| Robotics ML | ₹25–45 L | ₹45–80 L | ₹80 L–1.6 Cr |
Southeast Asia runs differently: Singapore is typically 2–3× Indian levels for equivalent seniority, Malaysia roughly 1.3–1.8×, with Thailand and Indonesia closer to Indian bands on the manufacturing-automation side.
Add recruiting fees — 15–25% of first-year salary in India, or 8.33% with us — and note that these searches take longer than general software: six to twelve weeks is normal, before a 60–90 day notice period.
India or Southeast Asia?
Both, usually, and for different reasons.
India has the larger pool, concentrated in Bengaluru, Pune, Hyderabad and Chennai, with real depth in embedded, controls and industrial robotics, and a growing perception and autonomy community.
Singapore has concentrated hardware, logistics-robotics and research talent, and is where a lot of regional robotics headquarters sit. Malaysia is strong on manufacturing automation and hardware engineering. Thailand and Indonesia are growing quickly on the industrial side.
For a warehouse or mobile-robotics build, running India and Southeast Asia in parallel reliably produces a better shortlist than either alone — and the notice-period and relocation mechanics differ enough between them that they need to be worked as separate searches rather than one.
How to assess properly
Keyword screening on "ROS" or "SLAM" tells you close to nothing. The signal is in failure.
Ask what broke. Robotics fails physically, and the gap between someone who has shipped and someone who has only prototyped appears immediately in how they describe it — calibration drift in the field, a sensor that behaved differently in rain, the edge case that only showed up on real hardware, the distance between simulation and reality.
Then ask:
- "Walk me through a system you took from bench to deployment. Who else touched it?"
- "What in that stack would you rebuild, and what would break if you did?"
- "Tell me about something that worked in sim and failed on the robot."
A candidate who has genuinely shipped answers these with specifics and some visible scar tissue. A candidate who has not gives you architecture diagrams.
What we do here
Robotics and hardware is one of our deepest verticals, and one we deliberately keep separate from general software recruiting because the networks do not overlap.
We work perception, SLAM, controls, embedded and robotics ML searches across India, Singapore, Malaysia, Thailand and Indonesia, for construction robotics, warehouse robotics, mobile robots, robot training data and autonomy teams. Our recruiters in this vertical are specialists who have placed these exact seats before — the pool is small enough that reach depends entirely on already being inside those circles.
One recent search: a senior perception engineer role, open four months, already worked by two internal recruiters and an agency. Agents scanned 1,916 profiles and surfaced 38 worth approaching — 31 of them employed and not looking. Fourteen replied. Four were sent with recordings, rubric scores and written cases. The offer signed on day nineteen. The person who signed was a perception lead at a company building the same thing, seven years in, not on a single job board.
Fee is 8.33% of first-year salary, invoiced on the day they join. No retainer, non-exclusive, 90-day replacement guarantee.
Robotics or physical AI role that has been open too long? Send it to us — and if the problem is the brief rather than reach, we will tell you that instead.