Ideas made tangible.

Project	Year	Domain	Key fact	Stack
LLM-powered cobot manipulationSpeak to it; it finds, poses and grasps any CAD-onboarded object	2025	manipulation	3 armsone robot interface	GPT-4 · Whisper · CNOS · FoundationPose · ZeroMQ	Story →
Humanoid teleoperationGloves and trackers drive humanoid arms and hands. No headset.	2025	teleoperation	100 Hz+sensor stream	Isaac Lab · ROS 2 · SenseGlove Nova2 · HTC Vive · Docker	Story →
Lab automation skills on a UR5eDrawer, knob, button, lid, rack. Skills from AprilTags and YAML.	2026	manipulation	1.9 mmhand-eye error	UR5e · ur_rtde · AprilTag · RealSense · MQTT	Story →
Hand-eye calibration, three waysBoard, fixed camera, or known geometry. Measured, not assumed.	2025	manipulation	0.59 mmrepeatability	OpenCV · PyTorch · ChArUco · AprilTag · Franka FR3	Story →
FoundationPose, measuredRobot scans, a ChArUco world frame and HouseCat6D. When to trust it.	2024	perception	89.4HouseCat6D IoU25 mAP	FoundationPose · Franka FR3 · Zivid · RealSense · SAM	Story →
Training-free 6D pose (OnePose++)Scan on an ArUco board, pose from dense matches. Then fix rotation.	2023	perception	45°LoFTR rotation limit	OnePose++ · LoFTR · se2-LoFTR · COLMAP · ZeroMQ	Story →
6D pose from synthetic-only dataKeypoints + PnP for a cube and a gripper, trained on Blender renders	2022	perception	4×TensorRT speed-up	PyTorch · Blender · OpenCV PnP · TensorRT	Story →
Self-supervised depth completionMetric depth from a camera and sparse LiDAR, no dense labels	2022	perception	0dense depth labels	PyTorch Lightning · PackNet-SfM · A2D2 · TensorRT	Story →
Automotive perception on embedded GPUsSegmentation, detection, depth and rain detection, shipped to cars	2020	systems	0.0mIoU lost at INT8	TensorFlow · PyTorch · TensorRT · C++/CUDA · Drive PX	Story →

Lab notes

Benchmarks · tools · experiments

Nov 2025Cosmos-Reason1 7B and Grounding DINO as ROS 2 nodes on OrintoolOn-robot VLM for a humanoid PoC
Feb 2025FoundationPose as a Docker + ZeroMQ pose servicetool~1.1 s / frame on a TITAN RTX
Dec 2024Pick-and-drop stress test, CAD-only graspingbenchmark302 / 342 picks (88.3%) in ~3 h
Jul 2024Finding a mirror-like metal ball with a tiny netexperimentLabels from ArUco + SAM, then OWLv2
Jun 2020TensorRT INT8 calibration on a real-time segmentation netbenchmarkINT8 0.5828 mIoU = FP32 0.5828
Dec 2019Rain and fog simulator with pixel-accurate rain maskstool22,000 synthetic training images
Oct 2019C++ Kalman library for ego-vehicle localisationtoolEKF + partial filters for GPS/IMU
Aug 2019Deep SORT tracker with swappable re-id featurestool3 appearance features, unit-tested
May 2019Ultrasound echoes to obstacle mapsexperimentFits training scenes, fails held-out
Jan 2019Per-class CRF tuning for segmentation post-processingtool200 TPE trials per class
Jul 2018Visual debugger for camera-to-map localisationtoolBuilt in one week, found 2 bugs
Dec 2017Faster capsule-network routing for traffic signsexperiment~10× faster training
Apr 2017Driver maneuver anticipation (MSc thesis)experiment~90% accuracy across maneuvers