arXiv subjects

Teng Cao

Publications and source records attributed to Teng Cao.

9 recordsLinked to original sources

arXiv2026

Explore More, Drift Less: Outcome-Only Reinforcement Learning Can Suffice for Long-Horizon Interactive Agents

Reinforcement learning is a natural way to post-train LLM agents for long-horizon interactive tasks judged only by end-of-task verification, yet a shared belief holds that outcome-only RL soon hits a ceiling on small open models. Recent work therefore compensates around the training with denser rewards, SFT priors, skill libraries, curated memory, or multi-agent orchestration. We argue the ceiling is an artifact of two failures of common practice. Signal starvation: group-relative RL with sparse outcome-only rewards yields a gradient only when a task's rollout group mixes successes and failures, so under-scaled exploration silences exactly the hardest, most instructive tasks. Policy drift: squeezing many updates out of a small task pool degrades the policy itself, as an unanchored objective lets the sampling distribution collapse exactly when saturation has already made informative groups rare. We present CANOPY (Coverage-ANchored On-PolicY RL), a minimalist protocol attacking both directly: scale same-task exploration until the natural signal reappears, keep every update on-policy, KL-anchored, and confined to the agent's own action tokens, then cash in an enlarged interaction budget at test time. On AppWorld, a long-horizon interactive coding benchmark, a Qwen3-14B policy trained with CANOPY through environment interaction alone--without task-specific supervision, auxiliary credit signals, or elaborate agent scaffolding--topped the public leaderboard (Feb. 2026; Test-Normal TGC 86.9, Test-Challenge 67.6), and the same design principles lift Qwen3.5-9B on SWE-bench Verified by 16.6 points. Agentic RL alone thus internalizes long-horizon capability directly into a small open model; we plan to release the complete training stack at https://github.com/AlibabaResearch/SignalCoverageRL.

cs.LG↗

arXiv2026

A New Paradigm for 3D Turbomachinery Design: Generative Diffusion Model Based Framework with Direct Geometry Encoding

The aerodynamic design of turbomachinery is critical to the performance of the overall energy system, yet it is challenging due to the complex non-linear flow physics and the presence of multiple-target design compromises. Denoising diffusion model, as one of the leading approaches in generative machine learning, has shown its advantages of high design solution accuracy and diversity in many engineering applications. In this study, we bring it to the 3D inverse design problem of turbomachinery, using centrifugal compressors as a classic representative, to demonstrate the new methodology for complex geometry designs. A diffusion model-centred design framework has been developed in this study. By specifying the desired design condition (mass flow rate and rotational speed) and targeted performance (pressure ratio and efficiency), the trained diffusion model returns directly the 3D compressor geometry that satisfies the condition inputs. Compared to traditional deterministic forward design approaches, the proposed method not only generates accurate geometry solutions to inverse design problems, but also enables effective exploration of the entire design space, providing a diverse set of candidate solutions. In addition, this paper presents the first study to directly train on 3D blade geometry coordinates rather than parametrised representations, demonstrating the feasibility of coordinate-based learning while enabling a highly flexible framework applicable to a wide range of designs. The trained diffusion model achieves excellent design capability, with solution accuracy up to 99% and unfeasible designs less than 1%. Furthermore, the solution diversity of the trained diffusion model is also quantitatively verified by means of comparing the distribution of the solution sets generated from the diffusion model and from direct sampling of physical parameters.

physics.flu-dyn↗

arXiv2026

Beyond Score Prediction: LLM-Based Essay Scoring and Feedback Generation via Reinforcement Learning with Rubric Rewards

Large language models (LLMs) have been widely applied to automated essay scoring (AES) and automated feedback generation (AFG). However, existing studies rely primarily on prompt engineering or supervised fine-tuning, while systematic research on reinforcement learning (RL) post-training and automated evaluation of feedback quality remains limited. We propose RLAES, a unified LLM framework that jointly optimizes essay scoring and feedback generation through RL. To make feedback quality measurable, interpretable, and usable for training, we introduce Rubric-based Feedback Evaluation (RFE), an essay-grounded feedback evaluation framework comprising 166 fine-grained binary rubric items and an LLM-as-judge. Building on RFE, we propose Adaptive Gated Feedback Optimization (AGFO), which activates rubric-based feedback rewards on demand during RL, reducing evaluation overhead while improving feedback quality. We also propose Adjacent Contrastive Reasoning (ACR) to improve ordinal score calibration by explicitly contrasting adjacent score levels. Experimental results show that the RFE framework captures essay-feedback consistency, exhibits strong pairwise discriminative power, and closely aligns with expert preferences. On the ASAP benchmark, RLAES-AGFO achieves the best scoring performance among LLM-based methods (QWK = 0.803), while maintaining feedback quality comparable to GPT-5.5 and avoiding the feedback degradation observed under score-only RL. Code and datasets are publicly available at https://github.com/hellomuyi/RLAES.

cs.CL↗

arXiv2026

Learning Explicit Behavioral Models with Adaptive Questions and World-Model Probes

Interactive agents trained only against task return can achieve high scores while failing to represent the mechanisms that make their actions succeed. This makes brittle behavior difficult to diagnose and limits adaptation when environment dynamics change. Existing LLM reflection and policy-code repair can revise behavior from failed trajectories, but questions and world-understanding tests are usually used only after training. We introduce an Explicit Symbolic Behavioral Model (ESBM), a trainable behavioral model that couples task performance with evidence-grounded question answering and executable mechanism prediction. An ESBM represents behavior through typed predicates, weighted rules, bounded options and mechanism memory; the mechanism layer predicts symbolic events, object changes, rewards and terminal consequences under action interventions. After each rollout, adaptive questions and active world-model probes convert score failures, QA errors and transition-prediction errors into constraints for local ESBM edits. Candidate models are selected by a multi-criterion rule that jointly evaluates task score, answerability and active world-model consistency. Under the tested Atari-style protocols, ESBM learns high-scoring policies while producing explicit answers and executable mechanism predictions, indicating that adaptive questions can serve as both training pressure and reusable benchmarks for mechanistic policy learning in this setting.

cs.LG↗

arXiv2026

Kintsugi: Learning Policies by Repairing Executable Knowledge Bases

Modern embodied agents achieve impressive performance, but their task knowledge is often stored in neural weights, latent state, or prompt-bound memory, making individual policy knowledge difficult to inspect, validate, recombine, and reuse. We introduce \textbf{Kintsugi}, a white-box policy-learning framework that treats embodied policy improvement as verifier-gated construction of a typed executable Knowledge Base (KB). Kintsugi represents task-level policy knowledge as composable typed entries -- predicates, operators, policy schemas, monitors, recovery rules, experience records, and goals -- and improves this artifact through localized typed edits induced from rollout evidence, rather than relying on test-time language-model reasoning. Between rollouts, a tool-constrained agentic editing loop diagnoses trajectory failures, localizes them to editable KB layers, and proposes candidate edits. A deterministic verification gate admits an edit only when the candidate type-checks, the resulting KB executes, and focused validation success or trajectory-health metrics improve without violating protected-regression checks. At inference, the accepted KB is executed by a deterministic symbolic executor with zero LLM calls. Across long-horizon text-agent benchmarks and representative object-centric manipulation settings, Kintsugi achieves strong endpoint performance while preserving inspectability, local editability, and verifier-gated deployment. These results suggest that embodied policy improvement can be organized around executable task knowledge.

cs.LG↗

arXiv2026

Prediction of Steady-State Flow through Porous Media Using Machine Learning Models

Solving flow through porous media is a crucial step in the topology optimisation of cold plates, a key component in modern thermal management. Traditional computational fluid dynamics (CFD) methods, while accurate, are often prohibitively expensive for large and complex geometries. In contrast, data-driven surrogate models provide a computationally efficient alternative, enabling rapid and reliable predictions. In this study, we develop a machine-learning framework for predicting steady-state flow through porous media governed by the Navier-Stokes-Brinkman equations. We implement and compare three model architectures-convolutional autoencoder (AE), U-Net, and Fourier Neural Operator (FNO)-evaluating their predictive performance. To enhance physics consistency, we incorporate physics-informed loss functions. Our results demonstrate that FNO outperforms AE and U-Net, achieving a mean squared error (MSE) as low as 0.0017 while providing speedups of up to 1000 times compared to CFD. Additionally, the mesh-invariant property of FNO emphasizes its suitability for topology optimisation tasks, where varying mesh resolutions are required. This study highlights the potential of machine learning to accelerate fluid flow predictions in porous media, offering a scalable alternative to traditional numerical methods.

physics.flu-dyn↗

arXiv2026

Diffusion Model Driven Airfoil Design: From Geometry Encoding to Practical Applications

Diffusion model, the state-of-the-art generative machine learning architecture, has shown promising results airfoil inverse designs. In this study, we implemented and trained a series of diffusion models on three different airfoil geometry data encoding formats -- principal component weights, ordered $x$-$y$ coordinates, and 2D signed distance functions (SDF) -- to generate 2D airfoils. By systematically comparing the performance of diffusion models trained on different data structures, it is found that for 2D airfoil design problems, the diffusion model performs the best when directly trained with coordinates. Training with latent space (PCA weights in this study) limits the model's design freedom, and decreases the training effectiveness. Although the 2D SDF data appears to result in the least performing model, it proves its feasibility in aerodynamic shape generation, paving the way towards 3D problems where SDF is more favored. This study also investigated deploying the diffusion model in practical engineering applications. A multi-target optimization procedure is proposed based on the stochastic nature of the diffusion process, which drastically simplifies the procedure compared to conventional methods. The extrapolation performance of the model is also investigated by tasking the model with both aerodynamic and flow condition labels that are extrapolated beyond the training set boundaries.

physics.flu-dyn↗

arXiv2025

STORM: Segment, Track, and Object Re-Localization from a Single Image

Accurate 6D pose estimation and tracking are core capabilities for physical AI systems, yet real-world deployment remains brittle and labor-intensive. Many pipelines rely on CAD models, manual masking, or per-object adaptation, and still fail under occlusion or fast motion without a principled way to recognize failure. We propose STORM, a unified framework for reference-conditioned 6D tracking that can operate from a single reference image, with minimal manual input and improved robustness. STORM combines: (i) Hierarchical Spatial Fusion Attention (HSFA), a task-driven reference-query fusion architecture that supports both single-reference and multi-reference conditioning and can optionally use vision-language semantic conditioning to resolve instance ambiguities; and (ii) a BCE-trained tracking verifier whose continuous compatibility logit is used as an energy-like score to detect drift and trigger automatic re-initialization. Experiments on LM-O and YCB-Video show that STORM improves annotation-free pose tracking accuracy over strong baselines and recovers reliably from severe occlusions and rapid viewpoint changes with minimal overhead.

cs.CV↗

arXiv2024

SuperPose: Improved 6D Pose Estimation with Robust Tracking and Mask-Free Initialization

We developed a robust solution for real-time 6D object detection in industrial applications by integrating FoundationPose, SAM2, and LightGlue, eliminating the need for retraining. Our approach addresses two key challenges: the requirement for an initial object mask in the first frame in FoundationPose and issues with tracking loss and automatic rotation for symmetric objects. The algorithm requires only a CAD model of the target object, with the user clicking on its location in the live feed during the initial setup. Once set, the algorithm automatically saves a reference image of the object and, in subsequent runs, employs LightGlue for feature matching between the object and the real-time scene, providing an initial prompt for detection. Tested on the YCB dataset and industrial components such as bleach cleanser and gears, the algorithm demonstrated reliable 6D detection and tracking. By integrating SAM2 and FoundationPose, we effectively mitigated common limitations such as the problem of tracking loss, ensuring continuous and accurate tracking under challenging conditions like occlusion or rapid movement.

cs.CV↗