While large language models (LLMs) have recently demonstrated strong potential in solving planning problems, there is a trade-off between flexibility and complexity. LLMs, as zero-shot planners themselves, are still not capable of directly generating valid plans for complex planning problems such as multi-constraint or long-horizon tasks. On the other hand, many frameworks aiming to solve complex planning problems often rely on task-specific preparatory efforts, such as task-specific in-context examples and pre-defined critics/verifiers, which limits their cross-task generalization capability. In this paper, we tackle these challenges by observing that the core of many planning problems lies in optimization problems: searching for the optimal solution (best plan) with goals subject to constraints (preconditions and effects of decisions). With LLMs' commonsense, reasoning, and programming capabilities, this opens up the possibilities of a universal LLM-based approach to planning problems. Inspired by this observation, we propose LLMFP, a general-purpose framework that leverages LLMs to capture key information from planning problems and formally formulate and solve them as optimization problems from scratch, with no task-specific examples needed. We apply LLMFP to 9 planning problems, ranging from multi-constraint decision making to multi-step planning problems, and demonstrate that LLMFP achieves on average 83.7\% and 86.8\% optimal rate across 9 tasks for GPT-4o and Claude 3.5 Sonnet, significantly outperforming the best baseline (direct planning with OpenAI o1-preview) with 37.6\% and 40.7\% improvements. We also validate components of LLMFP with ablation experiments and analyzed the underlying success and failure reasons.

Existing methods fail to effectively steer Large Language Models (LLMs) between textual reasoning and code generation, leaving symbolic computing capabilities underutilized. We introduce CodeSteer, an effective method for guiding LLM code/text generation. We construct a comprehensive benchmark SymBench comprising 37 symbolic tasks with adjustable complexity and also synthesize datasets of 12k multi-turn guidance/generation trajectories and 5.5k guidance comparison pairs. We fine-tune the Llama-3-8B model with a newly designed multi-turn supervised fine-tuning (SFT) and direct preference optimization (DPO). The resulting model, CodeSteerLLM, augmented with the proposed symbolic and self-answer checkers, effectively guides the code/text generation of larger models. Augmenting GPT-4o with CodeSteer raises its average performance score from 53.3 to 86.4, even outperforming the existing best LLM OpenAI o1 (82.7), o1-preview (74.8), and DeepSeek R1 (76.8) across all 37 tasks (28 seen, 9 unseen). Trained for GPT-4o, CodeSteer demonstrates superior generalizability, providing an average 41.8 performance boost on Claude, Mistral, and GPT-3.5. CodeSteer-guided LLMs fully harness symbolic computing to maintain strong performance on highly complex tasks. Models, Datasets, and Codes are available at https://github.com/yongchao98/CodeSteer-v1.0 and https://huggingface.co/yongchao98.

Recent works have shown great potential of Large Language Models (LLMs) in robot task and motion planning (TAMP). Current LLM approaches generate text- or code-based reasoning chains with sub-goals and action plans. However, they do not fully leverage LLMs’ symbolic computing and code generation capabilities. Many robot TAMP tasks involve complex optimization under multiple constraints, where pure textual reasoning is insufficient. While augmenting LLMs with predefined solvers and planners improves performance, it lacks generalization across tasks. Given LLMs’ growing coding proficiency, we enhance their TAMP capabilities by steering them to generate code as symbolic planners for optimization and constraint verification. Unlike prior work that uses code to interface with robot action modules or pre-designed planners, we steer LLMs to generate code as solvers, planners, and checkers for TAMP tasks requiring symbolic computing, while still leveraging textual reasoning to incorporate common sense. With a multi-round guidance and answer evolution framework, the proposed Code-as-Symbolic-Planner improves success rates by average 24.1% over best baseline methods across seven typical TAMP tasks and three popular LLMs. Code-as-Symbolic-Planner shows strong effectiveness and generalizability across discrete and continuous environments, 2D/3D simulations and real-world settings, as well as single- and multi-robot tasks with diverse requirements. See our project website†for prompts, videos, and code.

Yilun Hao, Yongchao Chen, Yang Zhang, Chuchu Fan. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Prompt optimization aims to find the best prompt to a large language model (LLM) for a given task. LLMs have been successfully used to help find and improve prompt candidates for single-step tasks. However, realistic tasks for agents are multi-step and introduce new challenges: (1) Prompt content is likely to be more extensive and complex, making it more difficult for LLMs to analyze errors, (2) the impact of an individual step is difficult to evaluate, and (3) different people may have varied preferences about task execution. While humans struggle to optimize prompts, they are good at providing feedback about LLM outputs; we therefore introduce a new LLM-driven discrete prompt optimization framework PROMST that incorporates human-designed feedback rules to automatically offer direct suggestions for improvement. We also use an extra learned heuristic model that predicts prompt performance to efficiently sample from prompt candidates. This approach significantly outperforms both human-engineered prompts and several other prompt optimization methods across 11 representative multi-step tasks (an average 10.6%-29.3% improvement to current best methods on five LLMs respectively). We believe our work can serve as a benchmark for automatic prompt optimization for LLM-driven multi-step tasks.

Multimodal demonstrations provide robots with an abundance of information to make sense of the world. However, such abundance may not always lead to good performance when it comes to learning sensorimotor control policies from human demonstrations. Extraneous data modalities can lead to state over-specification, where the state contains modalities that are not only useless for decision-making but also can change data distribution across environments. State over-specification leads to issues such as the learned policy not generalizing outside of the training data distribution. In this work, we propose Masked Imitation Learning (MIL) to address state over-specification by selectively using informative modalities. Specifically, we design a masked policy network with a binary mask to block certain modalities. We develop a bi-level optimization algorithm that learns this mask to accurately filter over-specified modalities. We demonstrate empirically that MIL outperforms baseline algorithms in simulated domains and effectively recovers the environment-invariant modalities on a multimodal dataset collected on a real robot. Videos and supplemental details are at: https://tinyurl.com/masked-il

Correspondence learning is a fundamental problem in robotics, which aims to learn a mapping between state, action pairs of agents of different dynamics or embodiments. However, current correspondence learning methods either leverage strictly paired data-which are often difficult to collect-or learn in an unsupervised fashion from unpaired data using regularization techniques such as cycle-consistency-which suffer from severe misalignment issues. We propose a weakly supervised correspondence learning approach that trades off between strong supervision over strictly paired data and unsupervised learning with a regularizer over unpaired data. Our idea is to leverage two types of weak supervision: i) temporal ordering of states and actions to reduce the compounding error, and ii) paired abstractions, instead of paired data, to alleviate the misalignment problem and learn a more accurate correspondence. The two types of weak supervision are easy to access in real-world applications, which simultaneously reduces the high cost of annotating strictly paired data and improves the quality of the learned correspondence. We show the videos of the experiments on our website.

Brain-inspired hyper-dimensional (HD) computing is a new computing paradigm emulating the neuron’s activity in high-dimensional space. The first step in HD computing is to map each data point into high-dimensional space (e.g., 10 000), which requires the computation of thousands of operations for each element of data in the original domain. Encoding alone takes about 80% of the execution time of training. In this article, we propose, ReHD, an entire rework of encoding, training, and inference in HD computing for a more hardware friendly implementation. ReHD includes a full binary encoding module for HD computing for energy-efficient and high-accuracy classification. Our encoding module based on random projection with a predictable memory access pattern can be efficiently implemented in hardware. ReHD is the first HD-based approach that provides data projection with a 1:1 ratio to the original data and enables all training/inference computation to be performed using binary hypervectors. After the optimizations ReHD adds to the encoding process, retraining and inference become the energy intensive part of HD computing. To resolve this, we additionally propose model quantization. Model quantization introduces a novel method of storing class hypervectors using$n$-bits, where$n$ranges from 1 to 32, rather than at full 32-bit precision, which allows for fine-grained tuning of the tradeoff between energy efficiency and accuracy. To further improve ReHD efficiency, we developed an online dimension reduction approach that removesinsignificanthypervector dimensions during training.

Brain-inspired Hyperdimensional (HD) computing is a novel and efficient computing paradigm which is more hardware-friendly than the traditional machine learning algorithms, however, the latest encoding and similarity checking schemes still require thousands of operations. To further reduce the hardware cost of HD computing, we present Stochastic-HD that combines the simplicity of operations in Stochastic Computing (SC) with the complex task solving capabilities of the latest HD computing algorithms. Stochastic-HD leverages deterministic SC, which uses structured input binary bitstreams instead of the traditional randomly generated bitstreams thus avoids expensive SC components like stochastic number generators. We also propose an in-memory hardware design for Stochastic-HD that exploits its high level of parallelism and robustness to approximation. Our hardware uses in-memory bitwise operations along with associative memory-like operations to enable a fast and energy-efficient implementation. With Stochastic-HD, we were able to reach a comparable accuracy with the Baseline-HD. As compared to the best PIM design for HD [1], Stochastic-HD is also 4.4% more accurate and 43.1× more energy-efficient.

Many classification problems in practice map each sample to more than one label - this is known as multi-label classification. In this work, we present Multi-label HD, an in 3D storage multi-label classification system that uses Hyperdimensional Computing (HD). Multi-label HD is the first HD system to support multi-label classification. We propose two different mappings of HD to Multi-label HD. The first, Power Set HD, transforms the multi-label problem into single-label classification by creating a new class for each label combination. The second, Multi-Model HD, creates a binary classification model for each possible label. Our evaluation shows that Multi-Model HD achieves, on average,$47.8\times$higher energy efficiency and$47.1\times$faster execution time while achieving 5% higher classification accuracy as state-of-the-art light-weight multi-label classifiers. Power Set HD achieves 13% higher accuracy than Multi-Model HD, but is$2\times$slower. Our 3D-flash acceleration further improves the energy efficiency of Multi-label HD training by$228\times$and reduces the latency by$610\times$vs training on a CPU.