🚀 We are excited to announce GreenAug (Green-screen Augmentation), a physical visual augmentation method for robot learning algorithms! GreenAug enables generalisation to unseen visually distinct locations (scenes). In collaboration with @TinkerSumit @yusufma555 @stepjamUK (1/6)

RT @stepjamUK: I've spent the last decade building robot infrastructure from scratch at research labs and companies. The waste of talent….

Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.

Will be at @corl_conf today!.

Super excited to be joining @wayve_ai as an ML Engineer in London! At Wayve, I will be working on end-to-end learning for autonomous driving.

For those unfamiliar with the elite sports community: Why wouldn’t athletes dope? Why wouldn’t they want a competitive advantage over something they have worked their entire life for?. Sports doping mostly exists because of corruption from sports committees and governments.

RT @vitalisvos19: Today we'll be presenting R&D at #RSS2024 during the Imitation Learning session at 8:30 am -- come by and visit our poste….

GreenAug code has been released!. We have the following implementations:.- GreenAugRandom.- GreenAugGenerative.- GreenAugMask.- GenerativeAugmentation - basically an implementation of ROSIE, with OS SOTA models (Grounding DINO, SAM, SD).

Best thing about unemployment is github green lights.

DMs are open! Also, feel free to email me at eugenetwc1@gmail.com.

RT @stepjamUK: 🚨Important update from our Robot Learning Lab in London. Following recent news, we’re moving on after a wonderful 2 years….T….

RT @mohito1905: Image-generation diffusion models can draw arbitrary visual-patterns. What if we finetune Stable Diffusion to 🖌️ draw joint….

RT @younggyoseo: Introducing CQN: Coarse-to-fine Q-Network, a value-based RL algorithm for continuous control🦾Initialized with 20~50 demons….

RT @nc__dev: 🚀 Looking for a benchmark for bi-manual mobile manipulation with nicely collected demonstrations? We are excited to release Bi….

⚙️ We advocate a shift in data collection practices. We propose that future research should use green screens for real-world demonstrations, followed by the application of GreenAug. This enables policy generalisation to new, visually distinct locations (scenes). 🌟 (6/6).

📊 In real-world experiments with over 850 training demos and 8.2k evaluation episodes, GreenAug outperforms no augmentation, standard computer vision augmentation, and previous generative augmentation methods (CACTI, GenAug, ROSIE). (5/6).

🔬 Our paper proposes GreenAug, which uses chroma keying to overlay background textures onto a green screen. We explore various GreenAug variants: random textures, generative model textures (e.g., Stable Diffusion), and training a masking network to isolate the background. (4/6).

🧵 Generalising vision-based robot learning policies to new environments is challenging and underexplored. Typically, data is collected in one location, and policies are trained and then deployed in the same location, which doesn't scale well. (3/6).

🌐 Website: 📄 Paper: .💻 Code: (2/6).

🔬 Our paper proposes GreenAug, which uses chroma keying to overlay background textures onto a green screen. We explore various GreenAug variants: random textures, generative model textures (e.g., Stable Diffusion), and training a masking network to isolate the background. (4/6).

🧵 Generalising vision-based robot learning policies to new environments is challenging and underexplored. Typically, data is collected in one location, and policies are trained and then deployed in the same location, which doesn't scale well. (3/6).