Jorge Mendez-Mendez
@jmendezm.bsky.social
Embodied lifelong learning (compositionality, RL, TAMP, robotics). Assistant Professor at Stony Brook ECE. Postdoc at MIT CSAIL, PhD from GRASP lab at Penn.
https://jorge-a-mendez.github.io
https://jorge-a-mendez.github.io
Maybe "correcting" wasn't the right word. The TAMP system tells the LLM what the mistakes were (e.g., a precondition wasn't met or an object couldn't be placed) and asks the LLM to generate a new plan, in a loop.
October 2, 2025 at 4:27 PM
Maybe "correcting" wasn't the right word. The TAMP system tells the LLM what the mistakes were (e.g., a precondition wasn't met or an object couldn't be placed) and asks the LLM to generate a new plan, in a loop.
The verdict? LLMs can't replace TAMP systems yet, especially for long-horizon, geometrically constrained tasks.
They are more promising as fast "idea generators", where a formal planner verifies their output.
Paper: arxiv.org/abs/2510.001...
Code & Data: github.com/jorge-a-mend...
They are more promising as fast "idea generators", where a formal planner verifies their output.
Paper: arxiv.org/abs/2510.001...
Code & Data: github.com/jorge-a-mend...
October 2, 2025 at 1:12 PM
The verdict? LLMs can't replace TAMP systems yet, especially for long-horizon, geometrically constrained tasks.
They are more promising as fast "idea generators", where a formal planner verifies their output.
Paper: arxiv.org/abs/2510.001...
Code & Data: github.com/jorge-a-mend...
They are more promising as fast "idea generators", where a formal planner verifies their output.
Paper: arxiv.org/abs/2510.001...
Code & Data: github.com/jorge-a-mend...
More counter-intuitive insight: giving the LLM more information can hurt performance!
When the prompt included geometric details, the LLM made more PDDL errors. The extra info "distracts" the model from the logical constraints. Are these systems sensitive to prompt engineering!
When the prompt included geometric details, the LLM made more PDDL errors. The extra info "distracts" the model from the logical constraints. Are these systems sensitive to prompt engineering!
October 2, 2025 at 1:12 PM
More counter-intuitive insight: giving the LLM more information can hurt performance!
When the prompt included geometric details, the LLM made more PDDL errors. The extra info "distracts" the model from the logical constraints. Are these systems sensitive to prompt engineering!
When the prompt included geometric details, the LLM made more PDDL errors. The extra info "distracts" the model from the logical constraints. Are these systems sensitive to prompt engineering!
Key finding: LLMs can solve many TAMP problems but lag behind engineered planners.
"Thinking" isn't always better! Non-reasoning LLMs outperformed reasoning ones.
Why? It's more efficient for the LLM generate plans quickly and have a formal TAMP system verify and correct them.
"Thinking" isn't always better! Non-reasoning LLMs outperformed reasoning ones.
Why? It's more efficient for the LLM generate plans quickly and have a formal TAMP system verify and correct them.
October 2, 2025 at 1:12 PM
Key finding: LLMs can solve many TAMP problems but lag behind engineered planners.
"Thinking" isn't always better! Non-reasoning LLMs outperformed reasoning ones.
Why? It's more efficient for the LLM generate plans quickly and have a formal TAMP system verify and correct them.
"Thinking" isn't always better! Non-reasoning LLMs outperformed reasoning ones.
Why? It's more efficient for the LLM generate plans quickly and have a formal TAMP system verify and correct them.
Broadcasting at its finest
December 16, 2024 at 5:44 AM
Broadcasting at its finest
I'm curious what the gap would be for getting this running with other platforms. Your thread seems to suggest it's fairly flexible and could be used with other 6-DoF mobile manipulators?
December 4, 2024 at 1:16 PM
I'm curious what the gap would be for getting this running with other platforms. Your thread seems to suggest it's fairly flexible and could be used with other 6-DoF mobile manipulators?
This looks awesome! Maybe I should consider a Stretch as my starter robot for my lab. 🤔
December 4, 2024 at 1:09 PM
This looks awesome! Maybe I should consider a Stretch as my starter robot for my lab. 🤔
We did spend a few weeks on deep RL. OOD and continual learning for sure will be there. What are some good reads for uncertainty quantification?
December 2, 2024 at 1:53 PM
We did spend a few weeks on deep RL. OOD and continual learning for sure will be there. What are some good reads for uncertainty quantification?
🙋♂️ I'm on the Island, if you're counting that.
November 30, 2024 at 5:31 PM
🙋♂️ I'm on the Island, if you're counting that.
To clarify, this starter pack is for lifelong learning of the *machine* variety.
November 26, 2024 at 2:26 PM
To clarify, this starter pack is for lifelong learning of the *machine* variety.
I work on embodied lifelong learning; would you mind adding me? Also, I made a related started pack for lifelong/continual learning: go.bsky.app/QXGY1iT. There may be some more folks there that are relevant!
November 25, 2024 at 7:38 PM
I work on embodied lifelong learning; would you mind adding me? Also, I made a related started pack for lifelong/continual learning: go.bsky.app/QXGY1iT. There may be some more folks there that are relevant!