> Markdown version of [/jobs/ext/4180628-software-development-engineer-ii](https://www.wearedevelopers.com/jobs/ext/4180628-software-development-engineer-ii). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Software Development Engineer II

- **Company:** Amazon.com, Inc.
- **Location:** Reading, MA, United States
- **Experience:** Experienced
- **Salary:** $143,700.0 - $194,400.0
- **Contract:** Permanent contract
- **Skills:** Computer Vision, C++ (Programming Language), Nvidia CUDA, Software Debugging, Software Design Documents, Middleware, Hardware-In-The-Loop Simulation, Python (Programming Language), Machine Learning, Object Detection, Model Compression, Software Engineering, Management of Software Versions, Graphics Processing Unit (GPU), Robot Operating System, Test-Driven Development (TDD), Deep Learning, Discretization, Real-time Inference, ONNX (Open Neural Network Exchange) Format, Triton Inference Server, Machine Learning Operations, TensorRT, NVIDIA Jetson
- **Published:** October 6, 2026
- **Apply:** https://www.indeed.com/viewjob?jk=9321d466e40f267c

## About the Role

* Knowledge of Python and/or C++ programming
 * 4+ years in CS, ME, EE or related field experience
 * Experience with robotics middleware like ROS2, ZCM, Zenoh, etc.
 * Experience developing or integrating perception, computer vision, SLAM, or navigation software for robots or other real-world embedded systems
 * Experience deploying and optimizing machine learning models for real-time inference on embedded or edge GPUs (e.g., NVIDIA Jetson), using tools such as TensorRT, ONNX Runtime, or CUDA
 * Experience with test driven development (TDD) in robotics software, * Master's degree or PhD in Robotics, CS, EE, or a related field, with a focus on computer vision or machine learning
 * Experience partnering with applied scientists to take research models from prototype to production, including improving their performance on real-world data
 * Experience with 3D perception and state estimation, such as stereo depth, visual or visual-inertial odometry, SLAM, mapping, or localization
 * Experience integrating perception into navigation or autonomy stacks (e.g., ROS 2, Nav2, occupancy or costmap generation)
 * Experience with model compression and quantization (post-training and quantization-aware, INT8/FP8, pruning, distillation) and GPU profiling tools such as NVIDIA Nsight Systems or Nsight Compute
 * Experience with multi-camera and multi-sensor pipelines (e.g., GMSL cameras, lidar) and serving multiple models on shared accelerators (e.g., Triton Inference Server)
 * Experience deploying models across heterogeneous accelerator platforms (GPU, NPU, DSP)
 * Experience building simulation-based or hardware-in-the-loop evaluation for perception systems
 * Excellent collaboration and communication skills, with the ability to work across science and engineering teams

## Description

As a Software Development Engineer on our on-robot software team, you will partner closely with applied scientists and engineers working on perception, localization and mapping (SLAM), and navigation to integrate their models and algorithms into the robot's onboard software, make them run in real time on embedded GPUs, and measurably improve their performance on real robot data. You will work at the boundary between science and engineering: turning research prototypes into fast, reliable, and observable on-robot components, and building the evaluation loops that show whether a change actually made the robot better., * Partner with perception, SLAM, and navigation scientists and engineers to integrate their models and algorithms (e.g., stereo depth, object detection and segmentation, visual-inertial odometry, mapping, and localization) into the on-robot software stack

 * Optimize deep learning models and computer vision pipelines for real-time inference on embedded GPUs, using techniques such as quantization, pruning, graph optimization, and inference-engine tuning
 * Profile and reduce end-to-end perception latency, throughput, and memory use from sensor input to model output, including when multiple models share the same GPU

 - Build efficient multi-camera and multi-sensor data paths, including zero-copy and hardware-accelerated pre- and post-processing, together with sensor-interface and model-serving owners
 * Build and own evaluation datasets, metrics, and regression gates (precision, recall, accuracy, latency) across offline, simulation, and hardware-in-the-loop testing, and use them with science partners to drive measurable accuracy improvements
 * Make perception components production-ready with versioning, configuration, health monitoring, and telemetry, and debug failures observed on real robots
 * Write design documents and experiment reports that clearly communicate tradeoffs, results, and limitations to science, engineering, and program stakeholders