We propose a Generative Adversarial Gumbel MCTS (GAG-MCTS) for abstract visual composition generation. It generates abstract visual concepts by fixed tangram pieces with the strengths of both tree search and GANs.

MCP-Universe is a benchmark for LLM agents that uses real-world MCP tools that simulate the real, challenging use cases with long-horizon reasoning and large tool spaces.

A pilot study on test-time scaling for GUI Agent using prolonged lookahead reasoning for action proposal and selection. It is coupled with a novel grounding model and achieves SOTA in the OSWorld leaderboard.

This paper introduces a Top-view Enhanced Perspective transformation (TEP) method that efficiently bridges robot and human perspectives for visual grounding tasks.

Auto-CEI pushes and estimates the limits of LLM reasoning capacities and aligns LLM's assertive and conservative response behaviours according to these limits for reliable reasoning.

We propose an easy-to-use framework that leverages sample and computational complexity from machine learning theory to analyze reasoning and planning problems and to design/optimize LLM-based reasoning methods.

We use Large Language Models as both the commonsense world model and the heuristic policy within Monte Carlo Tree Search. LLM's world model provides with MCTS a commonsense prior belief of states for reasoned decision-making. The LLM's heuristic policy guides the search to relevant parts of the tree, substantially reducing the search complexity.

We proposed ParaGon for language-conditioned object placing. ParaGon integrates a parsing algorithm into an end-to-end trainable neural network. It is data-efficient and generalizable for learning compositional instructions, and robust to noisy, ambiguous language inputs.

Risk-sensitive inverse reinforcement learning provides an general model to capture how human assess the distribution of a stochastic outcome when the true distribution is unknown (ambiguous). This work enables an RS-IRL learner to actively query expert demonstrations for faster risk envelope approximation.

Semantic SLAM projecting semantic meaning into 3D point clouds generated by ORB SLAM algorithm.