Perception/Autonomy Engineer (Degraded & Denied Environments)

About Kessari:

We are building autonomy systems that operate in GNSS-denied and degraded environments. This is not a research role. This is a real-world deployment on physical platforms.

We are specifically looking for engineers who think in terms of uncertainty, not clean inputs.

What this engineer must already think like:

• Vision is probabilistic, not deterministic
• Tracking is more important than detection (maintaining belief over time)
• Comfortable working with degraded inputs: low light, motion blur, packet loss, partial frames
• Design systems that continue operating when vision fails (fallback logic, sensor fusion)

Hard requirements (non-negotiable):

Strong experience with multi-object tracking:

Kalman filters, particle filters, JPDA, SORT/DeepSORT or similar

Sensor fusion experience:

• At minimum camera + IMU
• Bonus if experience with GNSS-denied navigation

Real-time systems experience:

• Edge compute, latency constraints, performance tradeoffs

Has deployed on real physical systems:

• UAVs, robotics, automotive, or defence systems
• Strong C++ (required), Python acceptable alongside

Nice to have (not required):

• SLAM / VIO (visual-inertial odometry)
• Tracking under occlusion and re-identification
• Experience in degraded or contested environments (e.g. EW, poor comms)
• Familiarity with PX4, ArduPilot, ROS

What we do NOT want:

• Pure machine learning engineers with no deployed systems
• Candidates focused only on object detection (YOLO-style pipelines)
• Academic researchers without real-world deployment experience

What they will be doing:

• Building and improving tracking systems that maintain target identity over time
• Designing sensor fusion pipelines robust to degraded inputs
• Working on real-time perception systems running on edge hardware
• Ensuring system reliability when inputs are unreliable or partially missing