RoboticsAI for Robotics

Home Robotics Engineering Platform

Perception, SLAM and embedded software for a US home-robotics company. Team of 3.

Key Details

Challenge	A US robotics company building home robots needed embedded software, perception and SLAM, model training data and integration support.
Solution	Embedded and robotics software covering perception and SLAM, AI and machine learning, robot data and training, plus integration and testing.
Technologies	C/C++, RTOS, Embedded Linux, Linux Kernel, U-Boot, Yocto, Buildroot, ROS, OpenCV, SLAM, Sensor fusion, PyTorch, JAX, CUDA, TensorRT

Technologies used

C/C++RTOSEmbedded LinuxYoctoROSOpenCVSLAMPyTorchJAXCUDATensorRT

Client

The client is a US-based robotics company building home robots for everyday physical tasks. The platform combines custom hardware, embedded software, perception, controls, autonomy and machine learning so a robot can work around people and pets. The reference names laundry and room reset as example tasks.

Product screenshots

Light grey engineering console showing a low camera view of a living room with labelled boxes for sofa, coffee table, armchair, pet bed, plant, door and a highlighted pair of slippers as an obstacle, a translucent green free-space overlay on the floor, small feature-point dots and an orange planned path curving around the obstacle, with side panels listing detections with distances, pose estimate, sensor topics and rates. — Perception view from the robot’s camera: detected furniture and floor obstacles with distances, free floor space, tracked SLAM features and the planned path around a pair of slippers.

Dark navy mapping screen with a top-down occupancy grid of a home floor plan split into coloured rooms, a cyan trajectory line, the robot icon with a field-of-view cone, magenta loop-closure markers and a dotted planned route to a goal, plus panels for localization confidence, sensor fusion inputs, map layers and a run log. — Map view: the visual SLAM map of the home with the robot’s localized pose, its trajectory, room segmentation and a sensor-fusion panel.

What we built

Embedded software, firmware and device drivers
Robot control, autonomy, middleware and system integration
3D perception, visual SLAM, sensor fusion and localization
Computer vision, robot learning and model training
Data pipelines, dataset processing and model evaluation
Hardware and software integration, simulation, diagnostics and automated testing

Engagement

Delivery is time and materials, run as Agile/Scrum, with 3 engineers over 10 months and is ongoing.

Project team: 3 engineers. Duration: 10 months, ongoing. Model: Time & materials, Agile/Scrum.

Services used

Embedded Software →AI for Robotics →