The posting
Robotics Research Engineer (L3) — Perceptyne
Hyderabad, India · Full-time · On-site with real robots
Build Robots with Hands That Work and Minds That Decide
Most robots today can move, do pick and place tasks. Very few can do dexterous work — plug a connector, seat a screw, mate a harness, handle a part that never sits quite where it did last time. That gap is the single biggest reason assembly lines still run on human hands.
Perceptyne builds industrial humanoid/semi-humanoid robots that close that gap: dual-arm, highly dexterous, contextually intelligent, and priced appropriately to actually deploy. Our robots are going into electronics and automotive assembly lines, e-commerce packaging, and more
This role involves defining and modifying architectures. The architecture you are part of setting now is the architecture a growing team builds on for years.
The shape of this role
You'll start as a senior individual contributor — hands on the code, hands on the robot. As the team grows around you, you'll be expected to step up: lead a small team, review others' work, own technical direction for your area, and coordinate across mechanical, embedded, and deployment teams.
What you'll actually do
- Co-Own the architecture. You will be part of defining how the robot's software stack is structured — module boundaries, interfaces, threading and real-time design, data flow between perception, planning, and control, configuration and versioning across robot variants. You'll be part of making the calls that decide whether shipping robot #50 is routine or painful, and you'll be able to defend those calls from first principles.
- Design motion planning and control algorithms for dual-arm dexterous manipulation, including contact-rich tasks where "plan a collision-free path" isn't enough.
- Build software that scales from one lab robot to a fleet in customer plants — reliable, observable, maintainable by people who didn't write it.
- Squeeze real-time performance out of the stack: multi-threaded C++, tight loops, latency budgets you can measure and defend.
- Prototype hypotheses in simulation, then prove or kill them on hardware. The robot is the referee.
- Work across the full system — perception, planning, manipulation, control, AI model training and deployment, tooling and UI, hardware bring-up, on-site deployment.
- Debug problems that cross domains, where the fault could be in your solver, the driver, the calibration, the gripper, or the mechanics. Find it anyway.
- Translate state-of-the-art robotics and AI research into something testable in days, not quarters.
- Plan and deliver: break large efforts into milestones, estimate honestly, flag slips early, and drive work to completion.
What we're looking for
- Bachelor's or Master's in CS, Electrical, Electronics, Instrumentation, or Mechanical Engineering.
- 4+ years building software for real robots, with hands-on manipulator experience.
- Demonstrated software architecture ability — you've designed a non-trivial robotics or real-time system, not just contributed modules to one. You can explain your abstractions, your trade-offs, and what you'd do differently now.
- Excellent C++ and Python. Real multi-threaded, real-time production code.
- Deep hands-on ROS/ROS 2, plus MoveIt, Nav2, OMPL and similar frameworks.
- Strong fundamentals in robotic manipulation: forward/inverse kinematics, Jacobians, dynamics, trajectory generation, control schemes.
- Real command of the underlying math — linear algebra, transforms, coordinate frames, calculus — and genuine physical intuition. You should sanity-check a result on a whiteboard before you trust the plot.
- Hands-on with depth cameras, lidar, and the messy reality of sensor data.
Multidisciplinary inclination — this one matters. Our hardest bugs and best ideas live at the seams between disciplines. We want someone who reaches across them by instinct: reads a mechanical drawing without flinching, talks RS485 bus and firmware timing with the embedded team, understands what a vision pipeline can and can't give a planner, knows where a learned policy beats a hand-written controller and where it doesn't. You won't be expert in all five of mechanical, embedded, CV, AI, and controls — but you should be fluent enough to design across them and curious enough to get your hands dirty in any of them.
Project ownership. Strong understanding of how real project timelines work, and the instinct to plan ahead rather than react. You sequence dependencies before they bite, you know which 20% of the work carries the schedule risk, and you close things out.
Nice to have: GPU programming; classical computer vision (feature extraction, template matching, optical flow, OCR); modern AI architectures (transformers, CNNs, RL, self-supervised learning) and a clear view of what each is actually good for; edge/cloud deployment experience; prior technical leadership or mentoring.
How we work
At our core: we like to build things, we evangelize great ideas, and we stay grounded. We hold a few deliberate tensions:
- Deep experience, without losing first-principles thinking.
- A large vision, without losing attention to detail.
- Disruptive technology, shipped to market as fast as possible.
We hire people with a strong bias for action, who are resourceful under constraints, who communicate clearly across mechanical, electrical, and software boundaries, and who keep going at problems that looked unsolvable on day one.
