About the role
You will own the perception and control stack that lets our systems know where they are and where the target is — without GPS, and without a network. That means the visual odometry that keeps a UAV flying its mission through jamming, the geospatial solver that turns a video frame into a firing correction, and the closed-loop control that puts a gun on that solution.
Whether you are two years in or coming off a competition team that shipped something that actually drove, flew, or shot, we care about spatial-computing fundamentals and how fast you iterate on real hardware.
What you will own:
- Visual odometry & navigation: build and tune the algorithm that dead-reckons an aircraft when GPS is jammed or spoofed, holds course in low light, and hands back seamlessly when the signal returns.
- Geospatial solving: own the pixel → line-of-sight → DEM ray-cast → geodetic fix pipeline, including camera models, DEM sampling, and WGS-84/ECEF↔ENU frame handling, to produce target coordinates and firing corrections.
- Drift correction: extend dead reckoning with offline satellite map-matching to recover an absolute position fix with no GPS and no connectivity.
- Closed-loop motion control: bridge perception to actuation — feedback loops, trajectory generation, and visual servoing that drive motor-actuated positioning to sub-degree accuracy, plus onboard AI target tracking.
- Field debugging: write Python and C++ tooling to pull telemetry, rebuild failure scenes from mission logs and recorded video, and chase down tracking dropouts, control lag, glare, vibration, and low-light degradation — on real airframes and guns, at trial sites.
- Edge optimization: work with the embedded team to make all of the above run in real time on constrained edge compute inside a sub-200g envelope.
Must have
* 2+ years of robotics/perception software experience, or hands-on autonomy build experience from a serious competition team (autonomous racing, RoboMaster, UAV/field-robotics, Formula Student, CubeSat).
* Deep grounding in classical computer vision and 3D geometry — camera models, coordinate frames, quaternions and rotation matrices, and reasoning about what happens when two frames disagree.
* Proficiency in C++ and Python, with OpenCV in production, not just in a notebook.
* Practical grasp of feedback control: PID tuning, kinematic transforms, motion planning, and knowing why a loop oscillates on real hardware but not in sim.
* Comfort at the software-hardware boundary: deploying to a physical platform, capturing logs, and debugging failures you cannot reproduce at your desk.
Nice to have
* MAVLink / ArduPilot or PX4 integration experience — talking to a flight controller, not just a simulator.
* GPS-denied navigation, VIO, or SLAM work on real airframes.
* Geospatial/GIS depth: DEM data, geodetic datums, georeferencing, satellite imagery registration.
* ROS2, and profiling/optimizing vision pipelines on constrained edge compute.
* Ballistics, fire control, or any defense, aerospace, or field-deployed autonomy background.
* B.Tech in Robotics, CS, Mechatronics, or Aerospace (helpful, not required — a strong build record beats it).
Notes
Location: Hyderabad, on-site (flight testing and hardware iteration require physical presence).
Given the defense nature of the work, Indian citizenship is required, and candidates should be comfortable with the security, confidentiality, and eventual clearance/vetting expectations that come with government and armed-forces trials.
Compensation is below-market cash + meaningful ESOP (4-year vest, 1-year cliff).
Deep-tech Work India