Can an agent orchestrating policies and tools create novel robot behaviors and learn from its experience? Yes, but learning is only effective for tasks that base policies don't trivially solve, and where success isn't rare either (~20%–50% baseline success rate). And it takes tens of trials per task, so the agent can separate causal success factors from spurious correlations.
Orchestration solved tasks that no single policy could (39.2% vs. 27.5% on RoboLab), and the agent's learned lessons transferred across tasks (23%→32% on an unseen task).
Though I don't believe orchestration of multiple policies will be the dominant deployment paradigm, I think it can provide value in data collection, RL exploration, or continual learning.
More in my blog post: ryanlindeborg.com/blog/robot…
Ryan Lindeborg retweeted
I'm attending #RSS2026 for a couple papers/talks and as an RSS Pioneer! Please reach out to chat and/or come to the following:
- Monday 9:40 AM: Spotlight pres on detecting side effects at the Rethinking Safety Workshop by @ryanlindeborg (sites.google.com/andrew.cmu.…)
- Tues 4-5PM: Come visit my RSS Pioneers Poster! @RSSPioneers
- Wed 3:15PM: Robometer Reward Fxn talk + poster after by @yigitkkorkmaz and @aliangdw (robometer.github.io/)
- Thurs 3:15PM: TMRL pre-training for post-training talk + poster after by @matthewh6_ (weirdlabuw.github.io/tmrl/)
- Friday 9:30AM: Giving an invited talk at SemRob Workshop (semrob.github.io/)
- Friday 5:30PM: Invited talk at the Diffusion Workshop (rss2026-diffusion-robot-lear…) on TMRL
Excited to be at #RSS2026 in Sydney! Come check out our work on predicting robot side effects at the Rethinking Safety workshop, and TiPToP at the FM4RoboPlan workshop!
Ryan Lindeborg retweeted
State-of-the-art robot policies often need hundreds of hours of data. What if we needed none?
Introducing TiPToP: a manipulation system that zero-shots open-world tasks from pixels and language using vision foundation models and GPU-parallelized Task and Motion Planning (TAMP).
Ryan Lindeborg retweeted
A reward model that works, zero-shot, across robots, tasks, and scenes?
Introducing Robometer: Scaling general-purpose robotic reward models with 1M+ trajectories.
Enables zero-shot: online/offline/model-based RL, data retrieval + IL, automatic failure detection, and more!
🧵 (1/12)
Ryan Lindeborg retweeted
Excited to have written the first blog post for USC’s new robotics blog!
The first blog post is by @Jesse_Y_Zhang on how LLMs can help robots adapt to solving tasks in new and unfamiliar settings.
rasc.usc.edu/blog/llms-adapt…
Ryan Lindeborg retweeted
Having humans annotate data to pre-train robots is expensive and time-consuming!
Introducing SPRINT:
A pre-training approach using LLMs and offline RL to equip robots w/ many language-annotated skills while minimizing human annotation effort!
URL: clvrai.github.io/sprint/
🧵👇
V excited for this - was just trying to get mujoco up and running on my box. Anything that can be done to streamline setup and use will have a high impact on research
We’ve acquired the MuJoCo physics simulator (mujoco.org) and are making it free for all, to support research everywhere. MuJoCo is a fast, powerful, easy-to-use, and soon to be open-source simulation tool, designed for robotics research: dpmd.ai/mujoco-blog