iiinova — where your digital self keeps evolving

Glen Berseth @glenberseth.bsky.social · 28/09/2026

#IROS2026 has 1,933 papers. I filtered them down to 3 reading lists totalling 130 papers on robot learning, VLAs, RL fine-tuning of robot policies, imitation learning, and world models, with session times, rooms, and related workshops. Happy reading! www.fracturedplane.com/blog/2026/09...

031

Glen Berseth @glenberseth.bsky.social · 15/09/2026

In the 1960s, Minsky's lab had a robot arm using computer vision to see object geometry, plan a path, and assemble a set of parts. Foundational models today are still largely solving this same class of problem, just far cheaper and faster to set up.

172

Glen Berseth @glenberseth.bsky.social · 08/09/2026

At this year's Montreal Robotics Summer School, teams programmed a Unitree Go1 to walk itself to dinner — sim2real RL, vision-based nav, AprilTags and all. Congrats to our Winning teams (5 & 6)! Full recap 👉 fracturedplane.com/blog/2026/08/31/MRSS2026.html #Robotics #ReinforcementLearning #Mila

010

Glen Berseth @glenberseth.bsky.social · 24/08/2026

This particular point is complex and important. New researchers are still learning, and LLMs are good at producing more, not better. It's like giving someone a big paintbrush and telling them to paint a wall. If an expert painter is given brushes, we get a mural instead of a quick, solid colour.

000

Glen Berseth @glenberseth.bsky.social · 18/08/2026

#RL_conference is full of great papers that are hard to choose from, so I created a list of the papers that I am most interested in reading from the conference. You can find that list here ;). rlj.cs.umass.edu/2026/2026issue…

492

Glen Berseth @glenberseth.bsky.social · 17/08/2026

Stability leads to scaling! Fantastic first talk by Marc Bellemare.

030

Glen Berseth @glenberseth.bsky.social · 03/08/2026

What a person can accomplish with LLMs and other recent tools is amazing. I am looking for new project ideas for a data science course to help people understand where LLMs can be used instead of traditional methods and how to use them for real industry projects. Any ideas?

000

Glen Berseth @glenberseth.bsky.social · 26/07/2026

The reason to learn this skill is so that people can then uses their well tuned Baloney detector for themselves on their own work to make sure they produce the highest quality they can.

140

Glen Berseth @glenberseth.bsky.social · 08/07/2026

Made a short (not really short) reading list of #ICML papers to keep up on robotics and RL. Check out the list here www.fracturedplane.com/blog/2026/07...

fracturedplane.com

| Glen Berseth

042

Glen Berseth @glenberseth.bsky.social · 07/07/2026

In Seoul for #ICML2026. Message me to catch up or talk about why #Reinforcementlearning will always improve model performance. We just need to make it even better.

020

Glen Berseth @glenberseth.bsky.social · 26/06/2026

This year, RLC includes a show at Cirque du Soleil! However, there aren't unlimited tickets; the venue is only so big. So register soon to guarantee your spot!

000

Glen Berseth @glenberseth.bsky.social · 23/06/2026

Paid LLMs were down this morning. Again. So I did what any prof does — I open-sourced my whole local LLM setup. Remote GPU via SSH, Ollama, OpenCode, benchmarking scripts. For private docs, planning, learning, and when the cloud flakes out. It's like a spare tire for AI.

110

Glen Berseth @glenberseth.bsky.social · 14/06/2026

This has been a wild week! I have been planning to take my lecture recordings from 3 years + my slides and turn them into a book on #RobotLearning! It seemed like the perfect 12 days to push! I got about halfway through before Fable was disrupted. Stay tuned for updates.

141

Glen Berseth @glenberseth.bsky.social · 03/06/2026

Compositional generalization is hard enough without being restricted to the limited availability of robotics data. In our new paper we use VLMs to autonomously decompose long-horizon demonstrations into diverse, language-labeled sub-tasks boosting VLA robustness!

131

Glen Berseth @glenberseth.bsky.social · 01/06/2026

Exciting news! I am delighted to share the wonderful news that I have been granted tenure! This milestone marks the completion of my time as an Assistant Professor, and I am profoundly grateful for the support, collaboration, and inspiration I have received from this incredible community.

150

Glen Berseth @glenberseth.bsky.social · 21/05/2026

Now accepting submissions for journal papers to appear in RLC! Want more face time for your work and to attend an awesome party with a great community? Submit here.

010

Glen Berseth @glenberseth.bsky.social · 19/05/2026

Still true. There is a lot more knowledge left to be learned.

110

Glen Berseth @glenberseth.bsky.social · 07/05/2026

Now that #NeurIPS is a few hours past, back to robotics. Where are people getting their bi manual arm mounts?

040

Glen Berseth @glenberseth.bsky.social · 25/04/2026

To learn about our #iclr2026 paper on learning representations to improve combinatorial generalization go find @danielblawson9 and Adriana at our poster on Sat: 9:30 AM – 12:00 PM EDT Pavilion 4 P4-#4512. self-pred-bc.github.io

self-pred-bc.github.io

Self-Predictive Representations for Combinatorial Generalization in Behavioral Cloning

010

Glen Berseth @glenberseth.bsky.social · 20/04/2026

After a busy start to the year, I have a new lecture to share. Generalist models have been making good progress, but #sim2real just works. In this lecture, I explain domain randomization, simulation limitations, and connections to generalization.

171

Glen Berseth @glenberseth.bsky.social · 19/03/2026

The sub-agent teams are getting very good. So good, I may have to start providing my students with management textbooks soon.

010

Glen Berseth @glenberseth.bsky.social · 01/03/2026

@obsidian.md + Claude = : 🔥 Having your AI agent help organize and process your thoughts and ideas all on your computer is amazing.

020

Glen Berseth @glenberseth.bsky.social · 26/02/2026

Just a few days left to submit your work to @rl-conference.bsky.social ! All attendees will get an awesome show from Cirque du Soleil.

021

Glen Berseth @glenberseth.bsky.social · 25/02/2026

🚀 Montreal #Robotics Summer School 2026 — Applications Now Open! 📅 Dates: August 2–7, 2026 📍 Location: @mila-quebec.bsky.social a (in person) 🤖 Program: 1 day of activities + 5 days of intensive Robotics & AI training (theory and hands-on)

2102

Glen Berseth @glenberseth.bsky.social · 22/02/2026

I am rerunning my class on robot learning this year, and I plan to push many code examples to help others get to the ugly details fast. One of these details is how BC gets off track as network sizes change. Blog and notebook below.

1141

Glen Berseth @glenberseth.bsky.social · 09/02/2026

I have updated my tutorial on making Vision Language Action models. This tutorial starts with a basic Transformer and walks people through the steps to transform it into a full VLA that uses PaliGemma as the pretrained VLM. Links below.

1144

Glen Berseth @glenberseth.bsky.social · 03/02/2026

I am looking forward to our first round of speakers tomorrow at @Mila_Quebec #worldmodels workshop. world-model-mila.github.io

240

Glen Berseth @glenberseth.bsky.social · 14/01/2026

How, after billions of dollars spent on code generation tool development they still can't reliably generate a working Dockerfile...

020

Glen Berseth @glenberseth.bsky.social · 09/01/2026

I am finally fully benefiting from making my lecture content in LaTeX. Creating new content powered by LLMs to make examples and translate my content to a webpage and a book is a breeze. Just need to figure out how to add references faster.

060

Reposted by Glen Berseth

Christian Wolf @chriswolfvision.bsky.social · 07/01/2026

Reinforcement Learning Conference (RLC) was added to the AI conference DL countdown. rl-conference.cc March 1: Abstract DL (AoE) March 5: Submission DL (AoE) Conference: Montreal, Quebec, Canada, August 16th -19th, 2026.

rl-conference.cc

RLC 2026

0164

Glen Berseth @glenberseth.bsky.social · 24/12/2025

Another exciting year for more RL! Submit your work to the RL conference and join us to talk a out how to make RL even better.

040

Glen Berseth @glenberseth.bsky.social · 06/12/2025

I'll be giving a talk tomorrow in the Embodied World Models for Decision Making workshop at #NeurIPS. I will present a number of recent works from my lab on combining foundational models with planning and RL. When: 11:30-noon Where: Level Room 30A-E

0110

Glen Berseth @glenberseth.bsky.social · 25/11/2025

#VisionLanguage models are increasingly used for a wide range of problems, but seem complex to build. I wrote some code and recorded a tutorial in my lab yesterday to help others demystify how to create these models. #keepbuilding

261

Glen Berseth @glenberseth.bsky.social · 19/11/2025

My lab is looking for new students who are very passionate about foundational models and planning/RL/robotics. Apply via Mila. I will also be at #NeurIPS to discuss research ideas and opportunities. See notes below for application advice.

2174

Glen Berseth @glenberseth.bsky.social · 31/10/2025

This not only seems illegal but is a breach of the social contract and trust in this research system. I can't see any good reason for breaking the confidentially of research grants and reference letters.

120

Glen Berseth @glenberseth.bsky.social · 28/10/2025

@pcastr.bsky.social is adding good clarity to the challenges of deep RL. Awesome stuff.

050

Glen Berseth @glenberseth.bsky.social · 25/10/2025

I have been seeing the subtleties in this race play out as destroying relationships. Families can fall apart because of the pressure to keep going in this race. Take care of yourselves everyone.

020

Glen Berseth @glenberseth.bsky.social · 19/10/2025

Does anyone in this universe know which human being accepts join requests for the #mlnews ml-news Google group? groups.google.com/g/ml-news. It says "This group is moderated and maintained by IMLS (www.machinelearning.org)." but that link just goes to the ICML webpage...

groups.google.com

Machine Learning News - Google Groups

010

Glen Berseth @glenberseth.bsky.social · 17/10/2025

Surprise/empowerment/etc may be the fundamental objectives living organisms optimize, however it is very difficult to optimize these objectives. I will be giving a talk at international worlshop on #activeinference on how foundational models can help improve these methods.

140

Glen Berseth @glenberseth.bsky.social · 15/10/2025

For those interested in joining my lab, submit your application via the Mila form. This year I am particularly interested in students with skills/interests in robotics, reinforcement learning and, foundational models which will push forward the abilities of real world agents.

060

Glen Berseth @glenberseth.bsky.social · 10/10/2025

I am at #COLM2025 today to talk about AI, LLMs and simulation in the social simulation workshop. Come find me, happy to chat about all things AI, embodiment, and simulation.

030

Glen Berseth @glenberseth.bsky.social · 02/10/2025

The same could be said for science.

010

Glen Berseth @glenberseth.bsky.social · 01/10/2025

GRPO is more like REINFORCE than PPO. 1) It does not train a critic (no need with small variance) 2) The SCORE FUNCTION (difficult to call this an advantage) is over a batch using the same initial prompt (similar to the vine sample method from TRPO)

110

Glen Berseth @glenberseth.bsky.social · 24/09/2025

On my way to South Korea for a week packed with robotics at the conference on Robot Learning, Humanoids2025, and the global forum on mechanical engineering.

020

Glen Berseth @glenberseth.bsky.social · 22/09/2025

One of the most common logical fallacies I see is "GPUs are cooking, therefore progress." I see people with 1/10th the compute get 10x more progress because... they have a more thorough plan. #Moretimethinkinglesstimeburning

140

Glen Berseth @glenberseth.bsky.social · 08/09/2025

I suggest going out and talking to real people. They provide a much richer signal.

160

Glen Berseth @glenberseth.bsky.social · 23/08/2025

VLAs offer an avenue for generalist robot policies; however, naively following the action predictions leads to brittle or unsafe behaviours. We introduce VLAPS, which integrates model-based search with pre-trained VLA policies to improve performance without additional training.

180

Glen Berseth @glenberseth.bsky.social · 21/08/2025

Efficiency may be the most important. If we can't make these tools economical, they will not last.

031

Glen Berseth @glenberseth.bsky.social · 20/08/2025

My lab at @montrealrobotics.bsky.social was honoured to present our recent work to @mark-carney.bsky.social and Even Solomon explaining how AI enables new robotics that will drive innovation in Canada. It was a pleasure getting into the details with a quick dive into deterministic policy gradients!

2130