Are world models the missing piece for truly dexterous tactile robots? 🤖
Great to join the @ARIA_research programme meeting on robot dexterity to discuss how predictive models perform in contact-rich settings.
#WorldModels#Robotics#ARIA@a2i_oxford@oxfordrobots
@ahmedkar_ Thanks! And I agree. Pixel-level prediction has been incredibly useful, but it’s often a poor proxy for relevance. What we’re really after is structure that reflects influence on behaviour, not fidelity of reconstruction.
A recurring theme in our recent work (e.g. SPARTAN): predictive world models become more interpretable when they help identify which parts of the input stream actually influence behaviour.
Learning what matters, not just what happens, is central.
https://t.co/TNzBoJ1URM
Very excited to share our new work - SPARTAN: A Sparse Transformer Learning Local Causation.
We develop a Transformer world model that learns local causal dependencies between entities, leading to improved adaptation efficiency and robustness with accurate prediction.
🧵
On my way to the 7th UK Manipulation W/s - a great way to start 2026! Over the years it’s become a premier UK meeting place for exchanging ideas across planning, control, and learning in robot manipulation.
If you’re attending, do come say hello.
https://t.co/eGQZYIUB55
Always look forward to CoRL - for the people, papers & workshops. Sadly have to miss it this year.
Delighted that @junjungoal & Alex Mitchell are there representing A2I - and @JankowskiJulius is presenting new work from our Amazon team! 🎉
See you next year!
NVIDIA’s £2bn pledge to UK AI startups is recognition that we don’t just use AI: we invent it, shape it & apply it.
As @UniofOxford researcher & co-founder of a successful AI startup, I was proud to be in the room. #AI#UKTech
🎓 Multiple faculty positions @oxengsci ! 🎓
We welcome applications from outstanding candidates in robotics, and especially if you are working in areas such as human-robot interaction, mechanical design, novel robotic sensor design and/or field robotics. Closing soon...🚀
Multiple faculty positions at University of Oxford in @oxengsci - Join Us!
Robotics - Computer Vision - Machine Learning
Faculty positions in Oxford are typically linked to a college.
Please repost!
Amongst my favourite research directions this year: understanding model complexity and its link to generalization and intelligence. Progress here could mean leaner models, versatile representations, and less reliance on data/energy. Excited that we’re off to the races on this!
I’m pleased to announce our work which studies complexity phase transitions in neural networks! We track the Kolmogorov complexity of networks as they “grok”, and find a characteristic rise and fall of complexity, corresponding to memorization followed by generalization.
🧵
2️⃣ RAINZ CDT (w/ @UKAEAofficial): Robot manipulation for net-zero energy systems (assembly/disassembly focus).
🗓 Deadline: 31 Jan 2025
👉 https://t.co/Ms8c7Rmx35
1️⃣ AIMS CDT (@Amazon-supported, w/ @j_foerst): Foundational work on learning & curating versatile world models.
🗓 Deadline: 29 Jan 2025
👉 https://t.co/PdCNB4Od29
How can a transformer uncover local causal dependencies in dynamic systems, from simulations to real-world data? 🤔
The answer: Hard attention + sparsity. But with a twist.
Meet SPARTAN: More causal. More efficient. Just as accurate.
#Robotics#ML#AI#CausalAI
Very excited to share our new work - SPARTAN: A Sparse Transformer Learning Local Causation.
We develop a Transformer world model that learns local causal dependencies between entities, leading to improved adaptation efficiency and robustness with accurate prediction.
🧵
Excited for #CoRL2024! Can’t wait to connect, learn, and share our latest on learned latent representations for quadruped locomotion. Let’s chat about structured world models, representations, and all the other groundbreaking work coming up! 🚀 #robotics
🚀 Join us to push the boundaries of AI and robotics, working on cutting-edge research in real-world robot learning. You'll focus on developing multimodal world models with impactful applications in collaborative manufacturing and social care.🤖 #robotlearning#GenerativeAI
On my way to #ICRA2024. Looking forward to Japan! Looking forward to seeing old friends and making new ones! And looking forward to presenting some of the work from @a2i_oxford and collaborators in Yokohama with @Jack_T_Collins, @junjungoal and @jannikzuern…
Delighted to be at #ICRA2024. Interested in effective sim-2-real transfer for world models (WeBT7-CC.6)? Or benchmarking for robot assembly (ThAT9-CC.3)? Or predicting lane graphs for autonomous driving (ThBT6-CC.2)? Come and see us to meet, discuss, or just hang-out...
Trajectory optimisation in high dimensional spaces is notoriously hard. What if you could leverage basic experience of what the system can do and let a diffusion model and vanilla sim guide you? Stunning work led by @junjungoal with @ShaohongZhong and @Jack_T_Collins@a2i_oxford
We introduce D-Cubed, a novel trajectory optimisation method using a latent diffusion model trained from a task-agnostic play dataset, including only representative hand motions, to solve dexterous deformable object manipulation tasks!
(1/N)