arXiv subjects

Kevin Qiu

Publications and source records attributed to Kevin Qiu.

7 recordsLinked to original sources

arXiv2026

Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning

Reinforcement learning (RL) has become central to LLM post-training, yet the methods that dominate current pipelines, PPO and GRPO, represent only a narrow slice of what RL offers. Understanding why these methods prevail, and what alternatives exist, requires a principled examination of the design decisions that underlie any RL algorithm. This survey organizes that examination around three stages of algorithm construction. We begin with MDP creation: how the reward function, state space, action space, termination condition, and discount factor are, or could be, defined for LLM training. We then turn to exploration, covering temperature sampling, entropy regularization, intrinsic motivation, tree search, and curriculum learning. Finally, we address learning along four classical RL dimensions: model-free versus model-based, value-based versus policy-based versus actor-critic, on-policy versus off-policy, and credit assignment, including both Monte Carlo methods, which rely on full return estimates, and bootstrapping methods, which update estimates using other learned predictions. Mapping the LLM literature onto this taxonomy reveals a strikingly non-uniform distribution of research effort. Critic-free policy gradients and Monte Carlo credit assignment are densely populated, while value-based methods, off-policy actor-critic training, and bootstrapping-based credit assignment remain largely unexplored despite well-established counterparts in classical RL. These gaps represent concrete opportunities for transferring proven RL techniques to LLM training. By making these gaps explicit alongside the methods that have proven effective, this survey offers researchers in both RL and LLMs a shared framework for understanding current practice and identifying promising directions for future work.

cs.LG↗

arXiv2026

Swim2Real: VLM-Guided System Identification for Sim-to-Real Transfer

We present Swim2Real, a pipeline that calibrates a 16-parameter robotic fish simulator from swimming videos using vision-language model (VLM) feedback, requiring no hand-designed search stages. Calibrating soft aquatic robots is particularly challenging because nonlinear fluid-structure coupling makes the parameter landscape chaotic, simplified fluid models introduce a persistent sim-to-real gap, and controlled aquatic experiments are difficult to reproduce. Prior work on this platform required three manually tailored stages to handle this complexity. The VLM compares simulated and real videos and proposes parameter updates. A backtracking line search then validates each step size, tripling the accept rate from 14% to 42% by recovering proposals where the direction is correct but the magnitude is too large. Swim2Real calibrates all 16 parameters simultaneously, most closely matching real fish velocities across all motor frequencies (MAE = 7.4 mm/s, 43% lower than the next-best method), with zero outlier seeds across five runs. Motor commands from the trained policy transfer to the physical fish at 50 Hz, completing the pipeline from swimming video to real-world deployment. Downstream RL policies swim 12% farther than those from BayesOpt-calibrated simulators and 90% farther than CMA-ES. These results demonstrate that VLM-guided calibration can close the sim-to-real gap for aquatic robots directly from video, enabling zero-shot RL transfer to physical swimmers without manual system identification, a step toward automated, general-purpose simulator tuning for underwater robotics.

cs.RO↗

arXiv2026

Vid2Sid: Videos Can Help Close the Sim2Real Gap

Calibrating a robot simulator's physics parameters (friction, damping, material stiffness) to match real hardware is often done by hand or with black-box optimizers that reduce error but cannot explain which physical discrepancies drive the error. When sensing is limited to external cameras, the problem is further compounded by perception noise and the absence of direct force or state measurements. We present Vid2Sid, a video-driven system identification pipeline that couples foundation-model perception with a VLM-in-the-loop optimizer that analyzes paired sim-real videos, diagnoses concrete mismatches, and proposes physics parameter updates with natural language rationales. We evaluate our approach on a tendon-actuated finger (rigid-body dynamics in MuJoCo) and a deformable continuum tentacle (soft-body dynamics in PyElastica). On sim2real holdout controls unseen during training, Vid2Sid achieves the best average rank across all settings, matching or exceeding black-box optimizers while uniquely providing interpretable reasoning at each iteration. Sim2sim validation confirms that Vid2Sid recovers ground-truth parameters most accurately (mean relative error under 13\% vs. 28--98\%), and ablation analysis reveals three calibration regimes. VLM-guided optimization excels when perception is clean and the simulator is expressive, while model-class limitations bound performance in more challenging settings.

cs.RO↗

arXiv2025

Debate2Create: Robot Co-design via Multi-Agent LLM Debate

We introduce Debate2Create (D2C), a multi-agent LLM framework that formulates robot co-design as structured, iterative debate grounded in physics-based evaluation. A design agent and control agent engage in a thesis-antithesis-synthesis loop, while criterion-specific LLM judges provide multi-objective feedback to steer exploration. Across five MuJoCo locomotion benchmarks, D2C achieves the highest default-normalized score among the evaluated LLM-based and black-box baselines, with gains up to 3.2x on Ant and nearly 9x on Swimmer. Iterative debate yields 18-35% gains over compute-matched zero-shot generation, and D2C-generated rewards transfer to default morphologies in 4/5 tasks. These results suggest that structured, simulator-grounded multi-agent interaction is a useful mechanism for joint morphology-reward optimization under a fixed-topology, per-candidate-RL protocol. Project page: debate2create.github.io.

cs.RO↗

arXiv2025

TAG-K: Tail-Averaged Greedy Kaczmarz for Computationally Efficient and Performant Online Inertial Parameter Estimation

Accurate online inertial parameter estimation is essential for adaptive robotic control, enabling real-time adjustment to payload changes, environmental interactions, and system wear. Traditional methods often struggle to track abrupt parameter shifts or incur high computational costs, limiting their effectiveness in dynamic environments and for computationally constrained robotic systems. We introduce TAG-K, a lightweight extension of the Kaczmarz method that combines greedy randomized row selection for rapid convergence with tail averaging for robustness under noise and inconsistency. This design enables fast, stable parameter adaptation while retaining the low per-iteration complexity inherent to the Kaczmarz framework. We evaluate TAG-K in synthetic benchmarks and quadrotor tracking tasks against RLS, KF, and other Kaczmarz variants. TAG-K achieves 1.5x-1.9x faster solve times on laptop-class CPUs and 4.8x-20.7x faster solve times on embedded microcontrollers. More importantly, these speedups are paired with improved robustness to measurement noise and a 25% reduction in estimation error, leading to nearly 2x better end-to-end tracking performance. Website, documentation, and code available at: https://a2r-lab.org/TAG-K/.

cs.RO↗

arXiv2025

Ground Awareness in Deep Learning for Large Outdoor Point Cloud Segmentation

This paper presents an analysis of utilizing elevation data to aid outdoor point cloud semantic segmentation through existing machine-learning networks in remote sensing, specifically in urban, built-up areas. In dense outdoor point clouds, the receptive field of a machine learning model may be too small to accurately determine the surroundings and context of a point. By computing Digital Terrain Models (DTMs) from the point clouds, we extract the relative elevation feature, which is the vertical distance from the terrain to a point. RandLA-Net is employed for efficient semantic segmentation of large-scale point clouds. We assess its performance across three diverse outdoor datasets captured with varying sensor technologies and sensor locations. Integration of relative elevation data leads to consistent performance improvements across all three datasets, most notably in the Hessigheim dataset, with an increase of 3.7 percentage points in average F1 score from 72.35% to 76.01%, by establishing long-range dependencies between ground and objects. We also explore additional local features such as planarity, normal vectors, and 2D features, but their efficacy varied based on the characteristics of the point cloud. Ultimately, this study underscores the important role of the non-local relative elevation feature for semantic segmentation of point clouds in remote sensing applications.

cs.CV↗

arXiv2024

RoboMorph: Evolving Robot Morphology using Large Language Models

We introduce RoboMorph, an automated approach for generating and optimizing modular robot designs using large language models (LLMs) and evolutionary algorithms. Each robot design is represented by a structured grammar, and we use LLMs to efficiently explore this design space. Traditionally, such exploration is time-consuming and computationally intensive. Using a best-shot prompting strategy combined with reinforcement learning (RL)-based control evaluation, RoboMorph iteratively refines robot designs within an evolutionary feedback loop. Across four terrain types, RoboMorph discovers diverse, terrain-specialized morphologies, including wheeled quadrupeds and hexapods, that match or outperform designs produced by Robogrammar's graph-search method. These results demonstrate that LLMs, when coupled with evolutionary selection, can serve as effective generative operators for automated robot design. Our project page and code are available at https://robomorph.github.io.

cs.LG↗