iiinova — where your digital self keeps evolving

Orr Krupnik @orrkrup.bsky.social · 14/02/2026

Prime example (found on the other platform) of why we should be careful with reward specification / alignment / guardrails / <enter your favorite AI safety topic here>. How much of this is human guided, and how much is just optimizing the “get PR merged” reward? github.com/matplotlib/m...

github.com

[PERF] Replace np.column_stack with np.vstack().T by crabby-rathbun · Pull Request #31132 · matplotlib/matplotlib

This PR addresses issue #31130 by replacing specific safe occurrences of np.column_stack with np.vstack().T for better performance. IMPORTANT: This is a more targeted fix than originally proposed. ...

140

Orr Krupnik @orrkrup.bsky.social · 10/02/2026

Take a look at our new paper! We improve sample efficiency and performance in off-policy RL by prioritizing experience with the semantic knowledge of a pre-trained VLM, and not even a very large one 🌍🤖📈🏆 Glad for the opportunity to work with @eladsharony.bsky.social and @tomjur.bsky.social !

000

Orr Krupnik @orrkrup.bsky.social · 18/07/2025

Are there any good robotics and/or RL podcasts still running in 2025? I used to enjoy The Robot Brains by Pieter Abbeel and TalkRL by @robinchauhan.bsky.social , but open to different styles too!

050

Orr Krupnik @orrkrup.bsky.social · 22/04/2025

I find this idea really neat - VLMs are great at describing scenes, but LLMs are better reasoners, so let's use text as an interim representation. Kind of reminiscent of the bitter lesson, only on a more "local" scale arxiv.org/abs/2503.15108

arxiv.org

VIPER: Visual Perception and Explainable Reasoning for Sequential Decision-Making

While Large Language Models (LLMs) excel at reasoning on text and Vision-Language Models (VLMs) are highly effective for visual perception, applying those models for visual instruction-based planning ...

010

Reposted by Orr Krupnik

Tal Daniel @taldaniel.bsky.social · 19/02/2025

Check out our new #ICLR2025 paper: EC-Diffuser leverages a novel Transformer-based diffusion denoiser to learn goal-conditioned multi-object manipulation policy from pixels!👇 Paper: www.arxiv.org/abs/2412.18907 Project page: sites.google.com/view/ec-diff... Code: github.com/carl-qi/EC-D...

121

Reposted by Orr Krupnik

Mirco Mutti @mircomutti.bsky.social · 13/12/2024

If interested on our take on addressing inverse RL in large state spaces, go to meet @filippo_lazzati and @alberto_metelli in the poster session 5 #NeurIPS2024 today (paper -> arxiv.org/abs/2406.03812)

152

Reposted by Orr Krupnik

Aviv Tamar @aviv-tamar.bsky.social · 25/11/2024

Want to learn / teach RL?   Check out new book draft: Reinforcement Learning - Foundations sites.google.com/view/rlfound... W/ Shie Mannor & Yishay Mansour This is a rigorous first course in RL, based on our teaching at TAU CS and Technion ECE.

415435

Orr Krupnik @orrkrup.bsky.social · 19/11/2024

Some papers really feel like a glimpse into the future! This one also serves as a powerful reminder that a lot of what we're focused on in the AI + robotics space is constrained by the hardware we have. arxiv.org/abs/2411.11192

arxiv.org

Robot Metabolism: Towards machines that can grow by consuming other machines

Biological lifeforms can heal, grow, adapt, and reproduce -- abilities essential for sustained survival and development. In contrast, robots today are primarily monolithic machines with limited abilit...

020