Action, language, and embodiment
Connected iCub learning, recurrent neural networks, action-language research, and reusable CPU/GPU experiment infrastructure.
Robotics and GPU computing
GPU-accelerated cognitive robotics on the iCub and beyond.
This work was about making embodied AI experiments large enough to matter. iCub, Aquila, CUDA, and the rover work all came from the same pressure: perception, action, language, timing, and computation had to work together in one loop, at a scale worth testing.
The visible objects are the iCub lab photographs, the bound PhD thesis, NVIDIA HQ and GTC records, Aquila source material, and ESA rover work.
Research line
Compute changed which questions were practical.
Embodied learning needed larger experiments; Aquila and CUDA made them more practical; NVIDIA and ESA showed the work in public; the Plymouth research community made it real.
Slow experiments shrink the question
Embodied learning needed more practical iteration across sensing, action, simulation, timing, and larger neural models.
Reusable experiments became possible
The work made cognitive-robotics experiments easier to run, teach, inspect, and accelerate.
Thesis, NVIDIA, ESA, Aquila, and videos
The public links include the Plymouth thesis, NVIDIA CUDA spotlight, ESA rover material, SourceForge Aquila, and robotics media.
Research work
What the robotics work made visible.
The work connected robot sensing, body control, recurrent neural networks, language grounding, simulation, and GPU acceleration. The important part was not one algorithm. It was the operating loop that made experiments possible.
iCub action-language learning connected perception to movement.
Aquila made CPU/GPU robotics experiments easier to inspect.
CUDA acceleration changed which models were practical.
ESA rover simulation tested autonomy under terrain, sensing, and navigation constraints.
Plymouth lab
The iCub work was a research community.
This photograph places the work inside the Plymouth iCub lab with supervisor Angelo Cangelosi, colleagues, student and co-author Barry Bentley, and the robot at the centre. The thesis, Aquila tools, action-language videos, rover work, and GPU acceleration came from that living research environment.
Source links
Primary source links.
Links and videos that show the robotics and GPU work rather than merely naming it.
Thesis
GPU Computing for Cognitive Robotics
NVIDIA HQ
Santa Clara presentation and CUDA spotlight
NVIDIA Santa Clara
Welcome record from the GPU robotics visit
NVIDIA GTC 2014
Presenting Aquila and CUDA robotics
Aquila
Open cognitive-robotics toolkit
ESA rover work
Evolution in Robotic Islands
NVIDIA keynote
Research mentioned at SC11
Robotics
University of Plymouth robotics
Rover simulation
Autonomous Mars rover progress
Active vision
Camera motion as part of perception
2011 talk
Integrating action and language in humanoid robots
NVIDIA
The Plymouth iCub/Aquila work appeared in NVIDIA’s early GPU-computing story.
The NVIDIA connection began before the later research placement: an invited Santa Clara HQ presentation on GPU-accelerated iCub work, documented by the NVIDIA welcome sign, followed by the CUDA spotlight, the GTC poster on action acquisition, the 2014 GTC presentation, and the SC11 keynote mention. Together they place the Plymouth iCub/Aquila work inside the early GPU-computing period before deep learning made that hardware shift familiar.
GTC 2014
Presenting the GPU robotics toolkit.
This photograph shows the public presentation stage for the same line of work: Aquila, CUDA acceleration, iCub cognitive robotics, and the practical tooling needed to make larger embodied-learning experiments possible.
Selected publications
The research path.
These entries stay visible because they support the robotics and GPU work.
GPU Computing for Cognitive Robotics
PhD thesis, University of Plymouth. The thesis that held together the iCub, action-language, vision, and GPU acceleration work.
Open thesisAquila 2.0: Software Architecture for Cognitive Robotics
Reusable CPU/GPU software architecture for cognitive robotics experiments.
Open publication recordGPU-accelerated action acquisition through MTRNN
NVIDIA GTC poster from the pre-CNN-boom GPU-compute era: iCub action acquisition with multiple time-scales recurrent neural networks, Aquila, CUDA, and measured GPU speedups.
Open GTC posterIntegrating action and language in humanoid robots
Robotika.SK/STU Bratislava talk page preserving the abstract for the iTALK/iCub action-language work.
Open Robotika.SK talk pageAn island-model framework for evolving neuro-controllers for planetary rover control
ESA-hosted paper on autonomous navigation, rover simulation, and neuro-controller optimization.
Open ESA paperDBLP, Scholar, and ResearchGate
Independent bibliographic and research-profile trails for additional publications and citation discovery.
Why it still matters
Compute changes the question.
When an experiment takes too long, the research question quietly shrinks. GPU acceleration and reusable tools widened the space of possible robotics experiments. That concern with practical constraints later reappears in edge cameras, synthetic scenes, and spatial calibration.
Follow the line into spatial intelligence