Physical AI Engineer
Summary
Physical AI Engineer at 7RD builds drone perception and autonomy software end to end: training YOLO vision models, optimizing them for embedded hardware, reinforcement learning, planning/obstacle-avoidance, Yocto-based deployment, and SIL/HIL simulation testing. Core stack: Python, C++, PyTorch, embedded Linux.
JOB
DESCRIPTION
Physical AI
Engineer
Edge
AI, Autonomy and Simulation
Location: | New Delhi |
Employment Type: | Full Time |
Experience Level: | 3+ Years |
Department: | Engineering / AI and Autonomy |
Role
Purpose
The engineer
will turn research and system requirements into perception and autonomy
software that performs reliably on real hardware. The position requires
practical ownership from data collection and model development through onboard
optimization, integration with drone software, simulation, and controlled
hardware testing.
Key
Responsibilities
● Vision models: Develop, train, fine tune, and evaluate YOLO based object detection
and tracking pipelines for drone imagery and video.
● Data and evaluation: Build labeling, augmentation, dataset versioning, error analysis, and
repeatable evaluation workflows across lighting, motion, altitude, and
environmental conditions.
● Model optimization: Apply quantization, pruning, model conversion, and runtime profiling
to meet latency, memory, compute, power, and thermal limits on target hardware.
● Reinforcement learning: Design training environments, reward functions, observations, actions,
constraints, and evaluation procedures for autonomous behavior.
● Maneuver algorithms: Develop and test planning, guidance, obstacle avoidance, trajectory
generation, and control support algorithms for complex maneuvers within defined
safety limits.
● Embedded deployment: Package inference services and supporting software for embedded
computers, including Yocto recipes, dependencies, startup, logging,
configuration, and updates.
● System integration: Connect perception and autonomy modules with sensors, camera streams,
telemetry, navigation estimates, autopilot interfaces, and ground tools.
● Simulation and testing: Set up realistic scenarios, run software in the loop and hardware in
the loop tests, inject faults and edge cases, and compare simulated performance
with recorded or physical tests.
● Software delivery: Write maintainable Python and C++ code, review changes, automate tests
and builds, document design decisions, and produce reproducible releases.
● Cross functional support: Work with firmware, hardware, controls, and
simulation engineers to define interfaces, review risks, reproduce issues, and
support controlled field trials.
Core
Technical Areas
Area | Expected Knowledge and
Application |
Computer vision | YOLO training and inference, detection
metrics, tracking, camera geometry, preprocessing, image quality, and failure
analysis. |
Machine learning | PyTorch or a comparable framework, dataset
management, experiment tracking, generalization checks, and reproducible
training. |
Edge optimization | INT8 or mixed precision quantization,
calibration, operator support, model export, profiling, and accuracy versus
latency tradeoffs. |
Reinforcement learning | Policies, environments, reward design,
exploration, constraints, offline evaluation, and transfer from simulation to
hardware. |
Autonomy algorithms | State estimation inputs, path and trajectory
planning, collision avoidance, maneuver constraints, guidance logic, and
autopilot integration. |
Embedded Linux | Yocto recipes and images, cross compilation,
services, containers or packages where appropriate, device access, logs, and
deployment. |
Simulation and test | Scenario design, sensor and dynamics models,
software in the loop, hardware in the loop, regression tests, telemetry
replay, and performance measurement. |
Software engineering | Python and C++, Git, code review, unit and
integration tests, CI, clear interfaces, observability, and release
documentation. |
Requirements
Required
Qualifications
● Bachelor's degree or equivalent practical
experience in Computer Science, Electronics, Robotics, Aerospace Engineering,
or a closely related discipline.
● At least 3 years of hands on experience
developing and evaluating computer vision or autonomy software, including a
YOLO family detector deployed to embedded hardware.
● Practical knowledge of quantization and runtime
profiling, with evidence of measuring model accuracy, latency, memory use, and
hardware resource limits.
● Experience developing algorithms for robotic
or drone autonomy and testing them in simulation.
● Working knowledge of reinforcement learning
concepts and the ability to define meaningful training and evaluation setups.
● Strong Python skills, working C++ skills, and
experience integrating software on embedded Linux, including Yocto based
builds.
● Ability to debug across models, software,
sensors, system timing, and onboard hardware using logs and repeatable tests.
Preferred
Experience
● Experience with ArduPilot or PX4, MAVLink, ROS
2, camera and IMU data, GNSS, and telemetry pipelines.
● Deployment with ONNX, TensorRT, OpenVINO,
TFLite, or hardware specific inference runtimes and accelerators.
● Synthetic data generation, domain
randomization, visual simulation, and methods for reducing the gap between
simulation and field performance.
● Development of software in the loop or
hardware in the loop test benches, automated scenario suites, and telemetry
replay tools.
● Experience with real time constraints, fault
handling, safe behavior when perception is uncertain, and controlled field
validation.
● AI or autonomy development for defence,
aerospace, robotics, industrial automation, or other reliability sensitive
systems.