arXiv subjects

Chuchu Fan

Publications and source records attributed to Chuchu Fan.

At least 37 records · Page 2Linked to original sources

arXiv2026

R1-Code-Interpreter: LLMs Reason with Code via Supervised and Multi-stage Reinforcement Learning

Practical guidance on training Large Language Models (LLMs) to leverage Code Interpreter across diverse tasks remains lacking. We present R1-Code-Interpreter, an extension of a text-only LLM trained via multi-turn supervised fine-tuning (SFT) and reinforcement learning (RL) to autonomously generate multiple code queries during step-by-step reasoning. Unlike prior RL + tool-use efforts focused on narrow domains such as math or retrieval, we curate 144 diverse reasoning and planning tasks and show that training a general-purpose Code Interpreter across them presents significant challenges due to task heterogeneity and scarcity of effective samples. To address this, we introduce a multi-stage curriculum learning approach that partitions training samples by measured improvement potential. The RL training prioritizes samples with higher potential and gradually shifts to lower-potential ones, increasing the average RL gains from merely +3.4% to +9.3% across Qwen-2.5 models (3/7/14B). Our final model, R1-CI-14B, improves average accuracy on the 37 test tasks from 44.1% to 72.4%, outperforming text-only GPT-4o (58.6%) and GPT-4o with Code Interpreter (70.9%). Notably, R1-CI-14B also exhibits emergent self-checking behavior through code generation. Datasets, Codes, and Models are available at https://github.com/yongchao98/R1-Code-Interpreter and https://huggingface.co/yongchao98.

cs.AI↗

arXiv2026

Solving Parameter-Robust Avoid Problems with Unknown Feasibility using Reinforcement Learning

Recent advances in deep reinforcement learning (RL) have achieved strong results on high-dimensional control tasks, but applying RL to reachability problems raises a fundamental mismatch: reachability seeks to maximize the set of states from which a system remains safe indefinitely, while RL optimizes expected returns over a user-specified distribution. This mismatch can result in policies that perform poorly on low-probability states that are still within the safe set. A natural alternative is to frame the problem as a robust optimization over a set of initial conditions that specify the initial state, dynamics and safe set, but whether this problem has a solution depends on the feasibility of the specified set, which is unknown a priori. We propose Feasibility-Guided Exploration (FGE), a method that simultaneously identifies a subset of feasible initial conditions under which a safe policy exists, and learns a policy to solve the reachability problem over this set of initial conditions. Empirical results demonstrate that FGE learns policies with over 50% more coverage than the best existing method for challenging initial conditions across tasks in the MuJoCo simulator and the Kinetix simulator with pixel observations.

cs.LG↗

arXiv2026

Heuristic Search as Language-Guided Program Optimization

Large Language Models (LLMs) have advanced Automated Heuristic Design (AHD) in combinatorial optimization (CO) in the past few years. However, existing discovery pipelines often require extensive manual trial-and-error or reliance on domain expertise to adapt to new or complex problems. This stems from tightly coupled internal mechanisms that limit systematic improvement of the LLM-driven design process. To address this challenge, we propose a structured framework for LLM-driven AHD that explicitly decomposes the heuristic discovery process into modular stages: a forward pass for evaluation, a backward pass for analytical feedback, and an update step for program refinement. This separation provides a clear abstraction for iterative refinement and enables principled improvements of individual components. We validate our framework across four diverse real-world CO domains, where it consistently outperforms baselines, achieving up to $0.17$ improvement in QYI on unseen test sets. Finally, we show that several popular AHD methods are restricted instantiations of our framework. By integrating them in our structured pipeline, we can upgrade the components modularly and significantly improve their performance.

cs.NE↗

arXiv2026

Discrete Adjoint Matching

Computation methods for solving entropy-regularized reward optimization -- a class of problems widely used for fine-tuning generative models -- have advanced rapidly. Among those, Adjoint Matching (AM, Domingo-Enrich et al., 2025) has proven highly effective in continuous state spaces with differentiable rewards. Transferring these practical successes to discrete generative modeling, however, remains particularly challenging and largely unexplored, mainly due to the drastic shift in generative model classes to discrete state spaces, which are nowhere differentiable. In this work, we propose Discrete Adjoint Matching (DAM) -- a discrete variant of AM for fine-tuning discrete generative models characterized by Continuous-Time Markov Chains, such as diffusion-based large language models. The core of DAM is the introduction of discrete adjoint-an estimator of the optimal solution to the original problem but formulated on discrete domains-from which standard matching frameworks can be applied. This is derived via a purely statistical standpoint, in contrast to the control-theoretic viewpoint in AM, thereby opening up new algorithmic opportunities for general adjoint-based estimators. We showcase DAM's effectiveness on synthetic and mathematical reasoning tasks.

stat.ML↗

arXiv2026

Curiosity is Knowledge: Self-Consistent Learning and No-Regret Optimization with Active Inference

Active inference (AIF) unifies exploration and exploitation by minimizing the Expected Free Energy (EFE), balancing epistemic value (information gain) and pragmatic value (task performance) through a curiosity coefficient. Yet it has been unclear when this balance yields both coherent learning and efficient decision-making: insufficient curiosity can drive myopic exploitation and prevent uncertainty resolution, while excessive curiosity can induce unnecessary exploration and regret. We establish the first theoretical guarantee for EFE-minimizing agents, showing that a single requirement--sufficient curiosity--simultaneously ensures self-consistent learning (Bayesian posterior consistency) and no-regret optimization (bounded cumulative regret). Our analysis characterizes how this mechanism depends on initial uncertainty, identifiability, and objective alignment, thereby connecting AIF to classical Bayesian experimental design and Bayesian optimization within one theoretical framework. We further translate these theories into practical design guidelines for tuning the epistemic-pragmatic trade-off in hybrid learning-optimization problems, validated through real-world experiments.

cs.LG↗

arXiv2026

ReFORM: Reflected Flows for On-support Offline RL via Noise Manipulation

Offline reinforcement learning (RL) aims to learn the optimal policy from a fixed dataset generated by behavior policies without additional environment interactions. One common challenge that arises in this setting is the out-of-distribution (OOD) error, which occurs when the policy leaves the training distribution. Prior methods penalize a statistical distance term to keep the policy close to the behavior policy, but this constrains policy improvement and may not completely prevent OOD actions. Another challenge is that the optimal policy distribution can be multimodal and difficult to represent. Recent works apply diffusion or flow policies to address this problem, but it is unclear how to avoid OOD errors while retaining policy expressiveness. We propose ReFORM, an offline RL method based on flow policies that enforces the less restrictive support constraint by construction. ReFORM learns a behavior cloning (BC) flow policy with a bounded source distribution to capture the support of the action distribution, then optimizes a reflected flow that generates bounded noise for the BC flow while keeping the support, to maximize the performance. Across 40 challenging tasks from the OGBench benchmark with datasets of varying quality and using a constant set of hyperparameters for all tasks, ReFORM dominates all baselines with hand-tuned hyperparameters on the performance profile curves.

cs.LG↗

arXiv2026

Maximizing Reach-Avoid Probabilities for Linear Stochastic Systems via Control Architectures

The maximization of reach-avoid probabilities for stochastic systems is a central topic in the control literature. Yet, the available methods are either restricted to low-dimensional systems or suffer from conservative approximations. To address these limitations, we propose control architectures that combine the flexibility of Markov Decision Processes with the scalability of Model Predictive Controllers. The Model Predictive Controller tracks reference signals while remaining agnostic to the stochasticity and reach-avoid objective. Instead, the reach-avoid probability is maximized by optimally updating the controller's reference online. To achieve this, the closed-loop system, consisting of the system and Model Predictive Controller, is abstracted as a Markov Decision Process in which a new reference can be chosen at every time-step. A feedback policy generating optimal references is then computed via Dynamic Programming. If the state space of the system is continuous, the Dynamic Programming algorithm must be executed on a finite system approximation. Modifications to the Model Predictive Controller enable a computationally efficient robustification of the Dynamic Programming algorithm to approximation errors, preserving bounds on the achieved reach-avoid probability. The approach is validated on a perturbed 12D quadcopter model in cluttered reach-avoid environments proving its flexibility and scalability.

math.OC↗

arXiv2026

Safety on the Fly: Constructing Robust Safety Filters via Policy Control Barrier Functions at Runtime

Control Barrier Functions (CBFs) have proven to be an effective tool for performing safe control synthesis for nonlinear systems. However, guaranteeing safety in the presence of disturbances and input constraints for high relative degree systems is a difficult problem. In this work, we propose the Robust Policy CBF (RPCBF), a practical approach for constructing robust CBF approximations online via the estimation of a value function. We establish conditions under which the approximation qualifies as a valid CBF and demonstrate the effectiveness of the RPCBF-safety filter in simulation on a variety of high relative degree input-constrained systems. Finally, we demonstrate the benefits of our method in compensating for model errors on a hardware quadcopter platform by treating the model errors as disturbances. Website including code: www.oswinso.xyz/rpcbf/

math.OC↗

arXiv2026

Parameter-Robust MPPI for Safe Online Learning of Unknown Parameters

Robots deployed in dynamic environments must remain safe even when key physical parameters are uncertain or change over time. We propose Parameter-Robust Model Predictive Path Integral (PRMPPI) control, a framework that integrates online parameter learning with probabilistic safety constraints. PRMPPI maintains a particle-based belief over parameters via Stein Variational Gradient Descent, evaluates safety constraints using Conformal Prediction, and optimizes both a nominal performance-driven and a safety-focused backup trajectory in parallel. This yields a controller that is cautious at first, improves performance as parameters are learned, and ensures safety throughout. Simulation and hardware experiments demonstrate higher success rates, lower tracking error, and more accurate parameter estimates than baselines.

cs.RO↗

arXiv2025

Joint Scheduling of DER under Demand Charges: Structure and Approximation

We study the joint scheduling of behind-the-meter distributed energy resources (DERs), including flexible loads, renewable generation, and battery energy storage systems, under net energy metering tariffs with demand charges. The problem is formulated as a stochastic dynamic program aimed at maximizing expected operational surplus while accounting for renewable generation uncertainty. We analytically characterize the optimal control policy and show that it admits a threshold-based structure. However, due to the strong temporal coupling of the storage and demand charge constraints, the number of conditional branches in the policy scales combinatorially with the scheduling horizon, as it requires a look-ahead over future states. To overcome the high computational complexity in the general formulation, an efficient approximation algorithm is proposed, which searches for the peak demand under a mildly relaxed problem. We show that the algorithm scales linearly with the scheduling horizon. Extensive simulations using two open-source datasets validate the proposed algorithm and compare its performance against different DER control strategies, including a reinforcement learning-based one. Under varying storage and tariff parameters, the results show that the proposed algorithm outperforms various benchmarks in achieving a relatively small solution gap compared to a theoretical upper bound.

eess.SY↗

arXiv2025

Structured Interfaces for Automated Reasoning with 3D Scene Graphs

In order to provide a robot with the ability to understand and react to a user's natural language inputs, the natural language must be connected to the robot's underlying representations of the world. Recently, large language models (LLMs) and 3D scene graphs (3DSGs) have become a popular choice for grounding natural language and representing the world. In this work, we address the challenge of using LLMs with 3DSGs to ground natural language. Existing methods encode the scene graph as serialized text within the LLM's context window, but this encoding does not scale to large or rich 3DSGs. Instead, we propose to use a form of Retrieval Augmented Generation to select a subset of the 3DSG relevant to the task. We encode a 3DSG in a graph database and provide a query language interface (Cypher) as a tool to the LLM with which it can retrieve relevant data for language grounding. We evaluate our approach on instruction following and scene question-answering tasks and compare against baseline context window and code generation methods. Our results show that using Cypher as an interface to 3D scene graphs scales significantly better to large, rich graphs on both local and cloud-based models. This leads to large performance improvements in grounded language tasks while also substantially reducing the token count of the scene graph content. A video supplement is available at https://www.youtube.com/watch?v=zY_YI9giZSA.

cs.CV↗

arXiv2025

AuDeRe: Automated Strategy Decision and Realization in Robot Planning and Control via LLMs

Recent advancements in large language models (LLMs) have shown significant promise in various domains, especially robotics. However, most prior LLM-based work in robotic applications either directly predicts waypoints or applies LLMs within fixed tool integration frameworks, offering limited flexibility in exploring and configuring solutions best suited to different tasks. In this work, we propose a framework that leverages LLMs to select appropriate planning and control strategies based on task descriptions, environmental constraints, and system dynamics. These strategies are then executed by calling the available comprehensive planning and control APIs. Our approach employs iterative LLM-based reasoning with performance feedback to refine the algorithm selection. We validate our approach through extensive experiments across tasks of varying complexity, from simple tracking to complex planning scenarios involving spatiotemporal constraints. The results demonstrate that using LLMs to determine planning and control strategies from natural language descriptions significantly enhances robotic autonomy while reducing the need for extensive manual tuning and expert knowledge. Furthermore, our framework maintains generalizability across different tasks and notably outperforms baseline methods that rely on LLMs for direct trajectory, control sequence, or code generation.

cs.RO↗

arXiv2025

TGPO: Temporal Grounded Policy Optimization for Signal Temporal Logic Tasks

Learning control policies for complex, long-horizon tasks is a central challenge in robotics and autonomous systems. Signal Temporal Logic (STL) offers a powerful and expressive language for specifying such tasks, but its non-Markovian nature and inherent sparse reward make it difficult to be solved via standard Reinforcement Learning (RL) algorithms. Prior RL approaches focus only on limited STL fragments or use STL robustness scores as sparse terminal rewards. In this paper, we propose TGPO, Temporal Grounded Policy Optimization, to solve general STL tasks. TGPO decomposes STL into timed subgoals and invariant constraints and provides a hierarchical framework to tackle the problem. The high-level component of TGPO proposes concrete time allocations for these subgoals, and the low-level time-conditioned policy learns to achieve the sequenced subgoals using a dense, stage-wise reward signal. During inference, we sample various time allocations and select the most promising assignment for the policy network to rollout the solution trajectory. To foster efficient policy learning for complex STL with multiple subgoals, we leverage the learned critic to guide the high-level temporal search via Metropolis-Hastings sampling, focusing exploration on temporally feasible solutions. We conduct experiments on five environments, ranging from low-dimensional navigation to manipulation, drone, and quadrupedal locomotion. Under a wide range of STL tasks, TGPO significantly outperforms state-of-the-art baselines (especially for high-dimensional and long-horizon cases), with an average of 31.6% improvement in task success rate compared to the best baseline. The code will be available at https://github.com/mengyuest/TGPO

cs.RO↗

arXiv2025

TUMIX: Multi-Agent Test-Time Scaling with Tool-Use Mixture

While integrating tools like Code Interpreter and Search has significantly enhanced Large Language Model (LLM) reasoning in models like ChatGPT Agent and Gemini-Pro, practical guidance on optimal tool use is lacking. The core challenge is effectively combining textual reasoning, coding, and search for diverse questions. In this paper, we propose Tool-Use Mixture (TUMIX), an ensemble framework that runs multiple agents in parallel, each employing distinct tool-use strategies and answer paths. Agents in TUMIX iteratively share and refine responses based on the question and previous answers. In experiments, TUMIX achieves significant gains over state-of-the-art tool-augmented and test-time scaling methods, delivering an average accuracy improvement of up to 3.55% over the best baseline on Gemini-2.5-Pro and Gemini-2.5-Flash across key reasoning benchmarks, with near-equal inference costs. We find that agent diversity and quality are crucial and can be enhanced by using LLMs to auto-optimize agent designs. Furthermore, TUMIX can halt refinement upon reaching sufficient confidence, preserving performance at only 49% of the inference cost. Further scaling can achieve higher performance, albeit at a greater cost.

cs.CL↗

arXiv2025

Hierarchical Multi-Agent Reinforcement Learning with Control Barrier Functions for Safety-Critical Autonomous Systems

We address the problem of safe policy learning in multi-agent safety-critical autonomous systems. In such systems, it is necessary for each agent to meet the safety requirements at all times while also cooperating with other agents to accomplish the task. Toward this end, we propose a safe Hierarchical Multi-Agent Reinforcement Learning (HMARL) approach based on Control Barrier Functions (CBFs). Our proposed hierarchical approach decomposes the overall reinforcement learning problem into two levels learning joint cooperative behavior at the higher level and learning safe individual behavior at the lower or agent level conditioned on the high-level policy. Specifically, we propose a skill-based HMARL-CBF algorithm in which the higher level problem involves learning a joint policy over the skills for all the agents and the lower-level problem involves learning policies to execute the skills safely with CBFs. We validate our approach on challenging environment scenarios whereby a large number of agents have to safely navigate through conflicting road networks. Compared with existing state of the art methods, our approach significantly improves the safety achieving near perfect (within 5%) success/safety rate while also improving performance across all the environments.

cs.LG↗

arXiv2025

Code-as-Symbolic-Planner: Foundation Model-Based Robot Planning via Symbolic Code Generation

Recent works have shown great potentials 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, 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 https://yongchao98.github.io/Code-Symbol-Planner/ for prompts, videos, and code.

cs.RO↗

arXiv2025

Learning Plasma Dynamics and Robust Rampdown Trajectories with Predict-First Experiments at TCV

The rampdown phase of a tokamak pulse is difficult to simulate and often exacerbates multiple plasma instabilities. To reduce the risk of disrupting operations, we leverage advances in Scientific Machine Learning (SciML) to combine physics with data-driven models, developing a neural state-space model (NSSM) that predicts plasma dynamics during Tokamak à Configuration Variable (TCV) rampdowns. The NSSM efficiently learns dynamics from a modest dataset of 311 pulses with only five pulses in a reactor-relevant high-performance regime. The NSSM is parallelized across uncertainties, and reinforcement learning (RL) is applied to design trajectories that avoid instability limits. High-performance experiments at TCV show statistically significant improvements in relevant metrics. A predict-first experiment, increasing plasma current by 20% from baseline, demonstrates the NSSM's ability to make small extrapolations. The developed approach paves the way for designing tokamak controls with robustness to considerable uncertainty and demonstrates the relevance of SciML for fusion experiments.

physics.plasm-ph↗

arXiv2025

Language-Grounded Hierarchical Planning and Execution with Multi-Robot 3D Scene Graphs

In this paper, we introduce a multi-robot system that integrates mapping, localization, and task and motion planning (TAMP) enabled by 3D scene graphs to execute complex instructions expressed in natural language. Our system builds a shared 3D scene graph incorporating an open-set object-based map, which is leveraged for multi-robot 3D scene graph fusion. This representation supports real-time, view-invariant relocalization (via the object-based map) and planning (via the 3D scene graph), allowing a team of robots to reason about their surroundings and execute complex tasks. Additionally, we introduce a planning approach that translates operator intent into Planning Domain Definition Language (PDDL) goals using a Large Language Model (LLM) by leveraging context from the shared 3D scene graph and robot capabilities. We provide an experimental assessment of the performance of our system on real-world tasks in large-scale, outdoor environments. A supplementary video is available at https://youtu.be/8xbGGOLfLAY.

cs.RO↗