Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agent LLMs. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME (ROME is Obviously an Agentic Model), an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-based Policy Alignment (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of the ALE infrastructure.

RollArt: Scaling Agentic RL Training via Disaggregated Infrastructure

Agentic Reinforcement Learning (RL) enables Large Language Models (LLMs) to perform autonomous decision-making and long-term planning. Unlike standard LLM post-training, agentic RL workloads are highly heterogeneous, combining compute-intensive prefill phases, bandwidth-bound decoding, and stateful, CPU-heavy environment simulations. We argue that efficient agentic RL training requires disaggregated infrastructure to leverage specialized, best-fit hardware. However, naive disaggregation introduces substantial synchronization overhead and resource underutilization due to the complex dependencies between stages. We present RollArc, a distributed system designed to maximize throughput for multi-task agentic RL on disaggregated infrastructure. RollArc is built on three core principles: (1) hardware-affinity workload mapping, which routes compute-bound and bandwidth-bound tasks to bestfit GPU devices, (2) fine-grained asynchrony, which manages execution at the trajectory level to mitigate resource bubbles, and (3) statefulness-aware computation, which offloads stateless components (e.g., reward models) to serverless infrastructure for elastic scaling. Our results demonstrate that RollArc effectively improves training throughput and achieves 1.35-2.05\(\times\) end-to-end training time reduction compared to monolithic and synchronous baselines. We also evaluate RollArc by training a hundreds-of-billions-parameter MoE model for Qoder product on an Alibaba cluster with more than 3,000 GPUs, further demonstrating RollArc scalability and robustness. The code is available at https://github.com/alibaba/ROLL.

Reinforcement Learning Optimization for Large-Scale Learning: An Efficient and User-Friendly Scaling Library

We introduce ROLL, an efficient, scalable, and user-friendly library designed for Reinforcement Learning Optimization for Large-scale Learning. ROLL caters to three primary user groups: tech pioneers aiming for cost-effective, fault-tolerant large-scale training, developers requiring flexible control over training workflows, and researchers seeking agile experimentation. ROLL is built upon several key modules to serve these user groups effectively. First, a single-controller architecture combined with an abstraction of the parallel worker simplifies the development of the training pipeline. Second, the parallel strategy and data transfer modules enable efficient and scalable training. Third, the rollout scheduler offers fine-grained management of each sample's lifecycle during the rollout stage. Fourth, the environment worker and reward worker support rapid and flexible experimentation with agentic RL algorithms and reward designs. Finally, AutoDeviceMapping allows users to assign resources to different models flexibly across various stages.

Physically Consistent Humanoid Loco-Manipulation using Latent Diffusion Models

This paper uses the capabilities of latent diffusion models (LDMs) to generate realistic RGB human-object interaction scenes to guide humanoid loco-manipulation planning. To do so, we extract from the generated images both the contact locations and robot configurations that are then used inside a whole-body trajectory optimization (TO) formulation to generate physically consistent trajectories for humanoids. We validate our full pipeline in simulation for different long-horizon loco-manipulation scenarios and perform an extensive analysis of the proposed contact and robot configuration extraction pipeline. Our results show that using the information extracted from LDMs, we can generate physically consistent trajectories that require long-horizon reasoning.

Baichuan-M1: Pushing the Medical Capability of Large Language Models

The current generation of large language models (LLMs) is typically designed for broad, general-purpose applications, while domain-specific LLMs, especially in vertical fields like medicine, remain relatively scarce. In particular, the development of highly efficient and practical LLMs for the medical domain is challenging due to the complexity of medical knowledge and the limited availability of high-quality data. To bridge this gap, we introduce Baichuan-M1, a series of large language models specifically optimized for medical applications. Unlike traditional approaches that simply continue pretraining on existing models or apply post-training to a general base model, Baichuan-M1 is trained from scratch with a dedicated focus on enhancing medical capabilities. Our model is trained on 20 trillion tokens and incorporates a range of effective training methods that strike a balance between general capabilities and medical expertise. As a result, Baichuan-M1 not only performs strongly across general domains such as mathematics and coding but also excels in specialized medical fields. We have open-sourced Baichuan-M1-14B, a mini version of our model, which can be accessed through the following links.

Trajectory Optimization under Contact Timing Uncertainties

Most interesting problems in robotics (e.g., locomotion and manipulation) are realized through intermittent contact with the environment. Due to the perception and modeling errors, assuming an exact time for establishing contact with the environment is unrealistic. On the other hand, handling uncertainties in contact timing is notoriously difficult as it gives rise to either handling uncertain complementarity systems or solving combinatorial optimization problems at run-time. This work presents a novel optimal control formulation to find robust control policies under contact timing uncertainties. Our main novelty lies in casting the stochastic problem to a deterministic optimization over the uncertainty set that ensures robustness criterion satisfaction of candidate pre-contact states and optimizes for contact-relevant objectives. This way, we only need to solve a manageable standard nonlinear programming problem without complementarity constraints or combinatorial explosion. Our simulation results on multiple simplified locomotion and manipulation tasks demonstrate the robustness of our uncertainty-aware formulation compared to the nominal optimal control formulation.

Arm-Constrained Curriculum Learning for Loco-Manipulation of the Wheel-Legged Robot

Incorporating a robotic manipulator into a wheel-legged robot enhances its agility and expands its potential for practical applications. However, the presence of potential instability and uncertainties presents additional challenges for control objectives. In this paper, we introduce an arm-constrained curriculum learning architecture to tackle the issues introduced by adding the manipulator. Firstly, we develop an arm-constrained reinforcement learning algorithm to ensure safety and stability in control performance. Additionally, to address discrepancies in reward settings between the arm and the base, we propose a reward-aware curriculum learning method. The policy is first trained in Isaac gym and transferred to the physical robot to do dynamic grasping tasks, including the door-opening task, fan-twitching task and the relay-baton-picking and following task. The results demonstrate that our proposed approach effectively controls the arm-equipped wheel-legged robot to master dynamic grasping skills, allowing it to chase and catch a moving object while in motion. Please refer to our website (https://acodedog.github.io/wheel-legged-loco-manipulation) for the code and supplemental videos.

Baichuan 2: Open Large-scale Language Models

Large language models (LLMs) have demonstrated remarkable performance on a variety of natural language tasks based on just a few examples of natural language instructions, reducing the need for extensive feature engineering. However, most powerful LLMs are closed-source or limited in their capability for languages other than English. In this technical report, we present Baichuan 2, a series of large-scale multilingual language models containing 7 billion and 13 billion parameters, trained from scratch, on 2.6 trillion tokens. Baichuan 2 matches or outperforms other open-source models of similar size on public benchmarks like MMLU, CMMLU, GSM8K, and HumanEval. Furthermore, Baichuan 2 excels in vertical domains such as medicine and law. We will release all pre-training model checkpoints to benefit the research community in better understanding the training dynamics of Baichuan 2.

A Robot Leg with Compact Variable Stiffness Joint based on Leaf-Spring Mechanism

The legged robots with variable stiffness actuators (VSAs) can achieve energy-efficient and versatile locomotion. However, equipping legged robots with VSAs in real-world application is usually restricted by (i) the redundant mechanical structure design, (ii) limited stiffness variation range and speed, and (iii) high energy consumption in stiffness modulation. In this paper, we present a novel Variable-Length Leaf-Spring Actuator (VLLSA) in legged robots that aims to address the aforementioned limitations. The design is based on leaf-spring mechanism and we improve the structural design to make the proposed VSA (i) compact and lightweight in mechanical structure, (ii) precise in theoretical modeling, and (iii) capable of modulating stiffness with wide range, fast speed, and low energy consumption. Hardware experiments validate that the legged robot equipped with the proposed VLLSA has compact structure, high dynamic performance and low energy consumption.

Design and Control of a Bio-inspired Wheeled Bipedal Robot

Wheeled bipedal robots have the capability to execute agile and versatile locomotion tasks in unknown terrains, with balancing being a key criteria in evaluating their dynamic performance. This paper focuses on enhancing the balancing performance of wheeled bipedal robots through innovations in both hardware and software aspects. A bio-inspired mechanical design, inspired by the human barbell squat, is proposed and implemented to achieve an efficient distribution of load onto the limb joints. This design improves knee torque joint efficiency and facilitates control over the distribution of the center of mass (CoM). Meanwhile, a customized balance model, namely the wheeled linear inverted pendulum (wLIP), is developed. The wLIP surpasses other alternatives by providing a more accurate estimation of wheeled robot dynamics while ensuring balancing stability. Experimental results demonstrate that the robot is capable of maintaining balance while manipulating pelvis states and CoM velocity; furthermore, it exhibits robustness against external disturbances and unknown terrains.