Vision-Language-Action (VLA) models have shown strong potential for general-purpose robotic manipulation, but their reliance on expert demonstrations limits their ability to learn from failures and perform self-corrections. Reinforcement learning (RL) addresses these through self-improving interactions with the physical environment, but suffers from high sample complexity on real robots. We introduce World-Model-based Policy Optimization (WMPO), a principled framework for on-policy VLA RL without interacting with the real environment. In contrast to widely used latent world models, WMPO focuses on pixel-based predictions that align the "imagined" trajectories with the VLA features pretrained with web-scale images. Crucially, WMPO enables the policy to perform on-policy GRPO that provides stronger performance than the often-used off-policy methods. Extensive experiments in both simulation and real-robot settings demonstrate that WMPO (i) substantially improves sample efficiency, (ii) achieves stronger overall performance, (iii) exhibits emergent behaviors such as self-correction, and (iv) demonstrates robust generalization and lifelong learning capabilities.

In this paper, we propose a second-order deterministic actor-critic framework in reinforcement learning that extends the classical deterministic policy gradient method to exploit curvature information of the performance function. Building on the concept of compatible function approximation for the critic, we introduce a quadratic critic that simultaneously preserves the true policy gradient and an approximation of the performance Hessian. A least-squares temporal difference learning scheme is then developed to estimate the quadratic critic parameters efficiently. This construction enables a quasi-Newton actor update using information learned by the critic, yielding faster convergence compared to first-order methods. The proposed approach is general and applicable to any differentiable policy class. Numerical examples demonstrate that the method achieves improved convergence and performance over standard deterministic actor-critic baselines.

This work will elaborate the fundamental principles of physical artificial intelligence (Physical AI) from a scientific and systemic perspective. The aim is to create a theoretical foundation that describes the physical embodiment, sensory perception, ability to act, learning processes, and context sensitivity of intelligent systems within a coherent framework. While classical AI approaches rely on symbolic processing and data driven models, Physical AI understands intelligence as an emergent phenomenon of real interaction between body, environment, and experience. The six fundamentals presented here are embodiment, sensory perception, motor action, learning, autonomy, and context sensitivity, and form the conceptual basis for designing and evaluating physically intelligent systems. Theoretically, it is shown that these six principles do not represent loose functional modules but rather act as a closed control loop in which energy, information, control, and context are in constant interaction. This circular interaction enables a system to generate meaning not from databases, but from physical experience, a paradigm shift that understands intelligence as an physical embodied process. Physical AI understands learning not as parameter adjustment, but as a change in the structural coupling between agents and the environment. To illustrate this, the theoretical model is explained using a practical scenario: An adaptive assistant robot supports patients in a rehabilitation clinic. This example illustrates that physical intelligence does not arise from abstract calculation, but from immediate, embodied experience. It shows how the six fundamentals interact in a real system: embodiment as a prerequisite, perception as input, movement as expression, learning as adaptation, autonomy as regulation, and context as orientation.

Many approaches to AI safety rely on inspecting model outputs or activations, yet certain risks are inherently undetectable by inspection alone. We propose a complementary, architecture-agnostic approach that enhances safety through the aggregation of multiple generative models, with the aggregated model inheriting its safety from the safest subset of a given size among them. Specifically, we present a consensus sampling algorithm that, given $k$ models and a prompt, achieves risk competitive with the average risk of the safest $s$ of the $k$ models, where $s$ is a chosen parameter, while abstaining when there is insufficient agreement between them. The approach leverages the models' ability to compute output probabilities, and we bound the probability of abstention when sufficiently many models are safe and exhibit adequate agreement. The algorithm is inspired by the provable copyright protection algorithm of Vyas et al. (2023). It requires some overlap among safe models, offers no protection when all models are unsafe, and may accumulate risk over repeated use. Nonetheless, our results provide a new, model-agnostic approach for AI safety by amplifying safety guarantees from an unknown subset of models within a collection to that of a single reliable model.

Four-Period Detailed Design Period 1: Topic Selection and Initial Exploration Period 2: Principle Analysis and Model Design Period 3: Model Construction and Refinement Period 4: "Historical Technology Expo" with presentations [Includes detailed student reflection prompts, extension activities, and troubleshooting guidance...] Base Model: Generic Outline Interdisciplinary Lesson Plan Design Learning Objectives: Help students understand how physics influences historical progress... Cultivate ability to analyze social factors behind technological development... Class Schedule: Four periods covering physics review, historical technologies, case study, and modern applications. Assessment: Class participation, group reports, reflection journals [Subsequent periods contain only high-level bullet points without actionable details...] 12 Qualitative Analysis This comparison reveals dramatic capability differences for complex generation tasks. The Base Model produces only a generic outline with vague bullet points--entirely insufficient for classroom use. Both AutoSynth and Expert-Designed models generate outstanding, comprehensive lesson plans with detailed objectives, granular activities, and sophisticated assessment schemes. The subtle differences reflect their optimization processes: AutoSynth emphasizes systematic difficulty coverage (likely from iterative refinement), while Expert-Designed showcases deep assessment design expertise. Both represent quantum leaps over baseline, validating that specialized workflows-- automated or manual--are essential for professional-grade content. This supports our quantitative findings (Table 1): while Au-toSynth achieves lower human preference (51 percent vs 96 percent), it produces genuinely high-quality outputs far superior to baseline capabilities.

Learning dexterous and agile policy for humanoid and dexterous hand control requires large-scale demonstrations, but collecting robot-specific data is prohibitively expensive. In contrast, abundant human motion data is readily available from motion capture, videos, and virtual reality, which could help address the data scarcity problem. However, due to the embodiment gap and missing dynamic information like force and torque, these demonstrations cannot be directly executed on robots. To bridge this gap, we propose Scalable Physics-Informed DExterous Retargeting (SPIDER), a physics-based retargeting framework to transform and augment kinematic-only human demonstrations to dynamically feasible robot trajectories at scale. Our key insight is that human demonstrations should provide global task structure and objective, while large-scale physics-based sampling with curriculum-style virtual contact guidance should refine trajectories to ensure dynamical feasibility and correct contact sequences. SPIDER scales across diverse 9 humanoid/dexterous hand embodiments and 6 datasets, improving success rates by 18% compared to standard sampling, while being 10X faster than reinforcement learning (RL) baselines, and enabling the generation of a 2.4M frames dynamic-feasible robot dataset for policy learning. As a universal physics-based retargeting method, SPIDER can work with diverse quality data and generate diverse and high-quality data to enable efficient policy learning with methods like RL.

We present CrochetBench, a benchmark for evaluating the ability of multimodal large language models to perform fine-grained, low-level procedural reasoning in the domain of crochet. Unlike prior benchmarks that focus on high-level description or visual question answering, CrochetBench shifts the emphasis from describing to doing: models are required to recognize stitches, select structurally appropriate instructions, and generate compilable crochet procedures. We adopt the CrochetPARADE DSL as our intermediate representation, enabling structural validation and functional evaluation via execution. The benchmark covers tasks including stitch classification, instruction grounding, and both natural language and image-to-DSL translation. Across all tasks, performance sharply declines as the evaluation shifts from surface-level similarity to executable correctness, exposing limitations in long-range symbolic reasoning and 3D-aware procedural synthesis. CrochetBench offers a new lens for assessing procedural competence in multimodal models and highlights the gap between surface-level understanding and executable precision in real-world creative domains. Code is available at https://github.com/Peiyu-Georgia-Li/crochetBench.

Reinforcement learning (RL) has demonstrated considerable potential for enhancing reasoning in large language models (LLMs). However, existing methods suffer from Gradient Starvation and Policy Degradation when training directly on samples with mixed difficulty. To mitigate this, prior approaches leverage Chain-of-Thought (CoT) data, but the construction of high-quality CoT annotations remains labor-intensive. Alternatively, curriculum learning strategies have been explored but frequently encounter challenges, such as difficulty mismatch, reliance on manual curriculum design, and catastrophic forgetting. To address these issues, we propose AdaCuRL, a Adaptive Curriculum Reinforcement Learning framework that integrates coarse-to-fine difficulty estimation with adaptive curriculum scheduling. This approach dynamically aligns data difficulty with model capability and incorporates a data revisitation mechanism to mitigate catastrophic forgetting. Furthermore, AdaCuRL employs adaptive reference and sparse KL strategies to prevent Policy Degradation. Extensive experiments across diverse reasoning benchmarks demonstrate that AdaCuRL consistently achieves significant performance improvements on both LLMs and MLLMs.

Planning in high-dimensional decision spaces is increasingly being studied through the lens of learned representations. Rather than training policies or value heads, we investigate whether planning can be carried out directly in an evaluation-aligned embedding space. We introduce SOLIS, which learns such a space using supervised contrastive learning. In this representation, outcome similarity is captured by proximity, and a single global advantage vector orients the space from losing to winning regions. Candidate actions are then ranked according to their alignment with this direction, reducing planning to vector operations in latent space. We demonstrate this approach in chess, where SOLIS uses only a shallow search guided by the learned embedding to reach competitive strength under constrained conditions. More broadly, our results suggest that evaluation-aligned latent planning offers a lightweight alternative to traditional dynamics models or policy learning.

As algorithms increasingly mediate competitive decision-making, their influence extends beyond individual outcomes to shaping strategic market dynamics. In two preregistered experiments, we examined how algorithmic advice affects human behavior in classic economic games with unique, non-collusive, and analytically traceable equilibria. In Experiment 1 (N = 107), participants played a Bertrand price competition with individualized or collective algorithmic recommendations. Initially, collusively upward-biased advice increased prices, particularly when individualized, but prices gradually converged toward equilibrium over the course of the experiment. However, participants avoided setting prices above the algorithm's recommendation throughout the experiment, suggesting that advice served as a soft upper bound for acceptable prices. In Experiment 2 (N = 129), participants played a Cournot quantity competition with equilibrium-aligned or strategically biased algorithmic recommendations. Here, individualized equilibrium advice supported stable convergence, whereas collusively downward-biased advice led to sustained underproduction and supracompetitive profits - hallmarks of tacit collusion. In both experiments, participants responded more strongly and consistently to individualized advice than collective advice, potentially due to greater perceived ownership of the former. These findings demonstrate that algorithmic advice can function as a strategic signal, shaping coordination even without explicit communication. The results echo real-world concerns about algorithmic collusion and underscore the need for careful design and oversight of algorithmic decision-support systems in competitive environments.