UniPred

Unifying Deep Predicate Invention with Foundation Models
Qianwei Wang1,2,∗, Bowen Li1,∗, Zhanpeng Luo1,3, Yifan Xu2, Alexander Gray4, Tom Silver5, Sebastian Scherer1, Katia Sycara1, Yaqi Xie1,†
1Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA
2Computer Science and Engineering Division, University of Michigan
3Department of Computer Science, University of Pittsburgh
4Centaur AI Institute
5Department of Electrical and Computer Engineering, Princeton University

∗ Equal Contribution. † Corresponding author.

Contact: bowenli2@andrew.cmu.edu.

The work was partly done when Qianwei Wang and Zhanpeng Luo were Robotics Institute Summer Scholars associated with the Advanced Agent Robotics Technology Lab, CMU.

Paper arXiv Code

Unifying Deep Predicate Invention with Foundation Models for Long Horizon Planning

UniPred performs long horizon planning with a learned neural symbolic world model

Abstract

Long horizon robotic tasks are hard due to contin uous state action spaces and sparse feedback. Symbolic world models help by decomposing tasks into discrete predicates that capture object properties and relations. Existing methods learn predicates either top down, by prompting foundation models without data grounding, or bottom up, from demonstrations without high level priors. We introduce UniPred, a bilevel learning framework that unifies both. UniPred uses large lan guage models (LLMs) to propose predicate effect distributions that supervise neural predicate learning from low level data, while learned feedback iteratively refines the LLM hypotheses. Leveraging strong visual foundation model features, UniPred learns robust predicate classifiers in cluttered scenes. We further propose a predicate evaluation method that supports symbolic models beyond STRIPS assumptions. Across five simulated and one real robot domains, UniPred achieves 2 ∼ 4× higher success rates than top down methods and 3 ∼ 4× faster learning than bottom up approaches, advancing scalable and flexible symbolic world modeling for robotics.

Demo Videos

Some failures UniPred can recover

UniPred continuously replans based on observation feedback, detects action failures during execution, and performs recovery replanning. Through a total of 13 steps, the system successfully completes the task.

Some failures cannot

Long horizon manipulation tasks are inherently challenging. Many unpredictable factors can lead to failure, including planning failures and accumulated execution errors at each action step.

BibTeX

@inproceedings{unipred2025,
  title     = {UniPred: Unifying Deep Predicate Invention with Foundation Models for Long Horizon Planning},
  author    = {TODO},
  booktitle = {TODO},
  year      = {2025}
}