Robotics + GPU
Action, language, and acceleration
PhD work connecting iCub humanoid learning, action structure, neural networks, CUDA, Aquila, and ESA rover-control research.
Robotics folderGPU computing · AI · cognitive robotics
Dr Martin Peniak
An early pioneer in applying GPU computing to cognitive robotics.
Before GPU deep learning became mainstream, his work used CUDA to scale recurrent neural networks for the iCub humanoid robot. Alongside that work, he developed evolutionary rover controllers in a University of Plymouth–ESA collaboration, presented cognitive-robotics applications at NVIDIA HQ in 2011, and joined NVIDIA as a research intern in Silicon Valley in 2012.
I created this archive to keep the life, the work, and the places connected: robotics, GPU computing, camera technology, spatial mapping, public talks, older writing, and a place built slowly over years.
TEDx Bratislava
My dream about AI.
TEDx Bratislava brought the route into public view: Plymouth, study, robotics, GPU computing, gratitude, and the dream of embodied AI.
Watch TEDxBegin here
Five anchors to begin.
2008 – University of Plymouth – awardRevell Research Systems Prize
2009-2010 – ESA ACT study – researcherEvolution in Robotic Islands
2011-2014 – NVIDIA / CUDA – research recordDeveloping Robots with CUDA
2014 – thesis – Plymouth – authorGPU Computing for Cognitive Robotics
2014 – public talk – TEDxBratislava – speakerMy dream about AI
Archive routes
Read the record in six routes.
Story gives the life context. Work holds the chapters. Talks keep the public voice. Milestones keep the objects. Sources keep the record. Writing preserves the old voice.
Work
The work as four records.
The chapters move from research robots to applied cameras, spatial maps, and a real place shaped slowly over time.
AI cameras
AI cameras and applied vision
Camera-as-computer prototypes, edge inference, synthetic worlds, and the pressure of making ideas work outside the lab.
AI cameras record
Spatial intelligence
From pixels to place
Multi-camera calibration, floorplane reasoning, uncertainty, topology, and the work of making observations belong to the same world.
Spatial recordTao
A place built slowly.
Tao records the long work of making rough land usable: water, paths, planting, structures, seasons, and care.
Over years, rough land became water, paths, planting, structures, and care.
Lineage
Robots, cameras, coordinates, place.
The tools changed from rover simulators and humanoid robots to camera systems, mapped spaces, and land. The test stayed physical: ideas had to work in labs, rooms, landscapes, and weather.
ESA rover
Autonomy and sensing against terrain, uncertainty, and planetary-robotics constraints.
iCub and Aquila
Action, language, neural dynamics, body constraints, and GPU-accelerated experiments.
NVIDIA and CUDA
A 2011 NVIDIA HQ presentation, 2012 research internship, CUDA Spotlight, and GTC records made the early GPU robotics work visible.
Applied cameras
Perception systems moved outside the lab into edge devices, workflows, and real-world pressure.
Spatial intelligence
Cameras become more useful when observations share coordinates, topology, and uncertainty.
Tao
Land, water, paths, structures, seasons, repair, and memory.