Robotics Software Developer

What a Robotics Software Developer does daily

• ROS 2 system development — architecting and developing robot software systems using the Robot Operating System 2 (ROS 2); ROS 2 node design (the fundamental unit of ROS 2 software — a process that performs computation and communicates via topics, services, and actions); defining ROS 2 messages and interfaces; composing robot behaviour from a graph of communicating nodes; using ROS 2's quality-of-service (QoS) settings for reliable vs best-effort communication; ROS 2 lifecycle management for managed state transitions; this is the core technical activity of most robotics software developers
• Perception system development — processing sensor data to build a model of the robot's environment; lidar point cloud processing (PCL — Point Cloud Library; 3D object detection; ground segmentation; obstacle mapping); camera image processing (OpenCV; stereo depth estimation; visual odometry); sensor fusion (combining lidar, camera, IMU, and GPS data into a coherent environmental model using Kalman filters or particle filters); object detection and semantic segmentation (deploying YOLOv8 or Detectron2 on embedded hardware for real-time scene understanding); these perception algorithms determine what a robot knows about its environment
• Simultaneous Localisation and Mapping (SLAM) — the robotics problem of building a map of an unknown environment while simultaneously determining the robot's location within it; 2D SLAM for indoor mobile robots (GMapping, Cartographer, Nav2 with AMCL); 3D SLAM for outdoor autonomous vehicles (LIO-SAM, KISS-ICP, LOAM); visual SLAM (ORB-SLAM3, OpenVINS); SLAM is the foundational capability that enables any robot to navigate in an unstructured environment without GPS
• Motion planning and control — computing how a robot should move to achieve a goal while avoiding obstacles; path planning for mobile robots (A*, RRT, RRT* in Nav2 or MoveIt); trajectory planning for robot arms (MoveIt 2 — the standard ROS 2 motion planning framework for manipulators; inverse kinematics; collision-aware path planning); velocity controllers (PID controllers for precise velocity tracking; model predictive control for constrained optimal motion); the robot's ability to move smoothly, safely, and efficiently depends on motion planning quality
• Real-time embedded software — writing safety-critical software that runs on microcontrollers (STM32, ESP32) or real-time operating systems (FreeRTOS, Zephyr); ROS 2's micro-ROS for embedded nodes; motor control firmware; sensor driver development; hardware abstraction layers; meeting hard real-time deadlines (the control loop must execute within a fixed time window regardless of system load); this is the lowest-level robotics software — where code directly controls physical actuators
• Robot simulation — building high-fidelity simulations for testing and development; Gazebo (the standard ROS simulation environment) for physics-based simulation of robots, sensors, and environments; Isaac Sim (NVIDIA) for photorealistic simulation with synthetic data generation for training perception models; simulation is essential because physical robot testing is slow, expensive, and risks hardware damage; "test in simulation first" is the standard robotics engineering practice
• Autonomous navigation — Nav2 (ROS 2's Navigation Stack 2) for mobile robot navigation; behaviour trees for modular robot behaviour (BehaviorTree.CPP — the standard C++ behaviour tree library used in ROS 2 navigation); global path planning (computing a complete path from start to goal); local path planning (dynamic obstacle avoidance during execution); recovery behaviours (what to do when the robot gets stuck); the Nav2 stack is the starting point for any ROS 2-based autonomous mobile robot
• Manipulation and grasping — programming robot arms to pick, place, and manipulate objects; grasp pose estimation (using point clouds or RGB-D data to compute where and how to grasp an object); motion planning for manipulators (MoveIt 2); force-torque control for compliant manipulation (applying appropriate force without damaging fragile objects); the core skill for industrial robotics applications (assembly, packaging, sorting, pick-and-place)
• UAV (drone) software development — flight controller integration (PX4, ArduPilot — the standard open-source UAV autopilot stacks); MAVROS (ROS 2 interface to MAVLink-based flight controllers); autonomous waypoint navigation; precision landing; payload control; drone swarm coordination; in Sri Lanka, agricultural drones (crop spraying, yield estimation) and surveillance UAVs are the primary application areas
• AI and machine learning integration — deploying trained ML models (PyTorch, TensorFlow) on robot hardware; NVIDIA TensorRT for high-performance GPU inference on embedded Jetson hardware; OpenVINO for Intel edge inference; integrating LLMs for robot instruction following (a rapidly growing area — telling a robot "pick up the red cup from the table" using natural language commands interpreted by an LLM); reinforcement learning for robot control (Isaac Lab / Isaac Gym for training manipulation policies in simulation)