Search

The alignment of large language models (LLMs) is critical for developing effective and safe language models. Traditional approaches focus on aligning models during the instruction tuning or reinforcement learning stages, referred to in this paper as `post alignment'. We argue that alignment during the pre-training phase, which we term `native alignment', warrants investigation. Native alignment aims to prevent unaligned content from the beginning, rather than relying on post-hoc processing. This approach leverages extensively aligned pre-training data to enhance the effectiveness and usability of pre-trained models. Our study specifically explores the application of native alignment in the context of Arabic LLMs. We conduct comprehensive experiments and ablation studies to evaluate the impact of native alignment on model performance and alignment stability. Additionally, we release open-source Arabic LLMs that demonstrate state-of-the-art performance on various benchmarks, providing significant benefits to the Arabic LLM community., Comment: Accepted to NeurIPS 2024 main conference. see https://github.com/FreedomIntelligence/AceGPT-v2

Enhancing the modular structure of existing systems has attracted substantial research interest, focusing on two main methods: (1) software modularization and (2) identifying design issues (e.g., smells) as refactoring opportunities. However, re-modularization solutions often require extensive modifications to the original modules, and the design issues identified are generally too coarse to guide refactoring strategies. Combining the above two methods, this paper introduces a novel concept, PairSmell, which exploits modularization to pinpoint design issues necessitating refactoring. We concentrate on a granular but fundamental aspect of modularity principles -- modular relation (MR), i.e., whether a pair of entities are separated or collocated. The main assumption is that, if the actual MR of a pair violates its `apt MR', i.e., an MR agreed on by multiple modularization tools (as raters), it can be deemed likely a flawed architectural decision that necessitates further examination. To quantify and evaluate PairSmell, we conduct an empirical study on 20 C/C++ and Java projects, using 4 established modularization tools to identify two forms of PairSmell: inapt separated pairs InSep and inapt collocated pairs InCol. Our study on 260,003 instances reveals that their architectural impacts are substantial: (1) on average, 14.60% and 20.44% of software entities are involved in InSep and InCol MRs respectively; (2) InSep pairs are associated with 190% more co-changes than properly separated pairs, while InCol pairs are associated with 35% fewer co-changes than properly collocated pairs, both indicating a successful identification of modular structures detrimental to software quality; and (3) both forms of PairSmell persist across software evolution., Comment: Accepted by 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE'25)

Running optimization across many parallel seeds leveraging GPU compute have relaxed the need for a good initialization, but this can fail if the problem is highly non-convex as all seeds could get stuck in local minima. One such setting is collision-free motion optimization for robot manipulation, where optimization converges quickly on easy problems but struggle in obstacle dense environments (e.g., a cluttered cabinet or table). In these situations, graph-based planning algorithms are used to obtain seeds, resulting in significant slowdowns. We propose DiffusionSeeder, a diffusion based approach that generates trajectories to seed motion optimization for rapid robot motion planning. DiffusionSeeder takes the initial depth image observation of the scene and generates high quality, multi-modal trajectories that are then fine-tuned with a few iterations of motion optimization. We integrate DiffusionSeeder to generate the seed trajectories for cuRobo, a GPU-accelerated motion optimization method, which results in 12x speed up on average, and 36x speed up for more complicated problems, while achieving 10% higher success rate in partially observed simulation environments. Our results show the effectiveness of using diverse solutions from a learned diffusion model. Physical experiments on a Franka robot demonstrate the sim2real transfer of DiffusionSeeder to the real robot, with an average success rate of 86% and planning time of 26ms, improving on cuRobo by 51% higher success rate while also being 2.5x faster.

Generative AI systems have shown impressive capabilities in creating text, code, and images. Inspired by the rich history of research in industrial ''Design for Assembly'', we introduce a novel problem: Generative Design-for-Robot-Assembly (GDfRA). The task is to generate an assembly based on a natural language prompt (e.g., ''giraffe'') and an image of available physical components, such as 3D-printed blocks. The output is an assembly, a spatial arrangement of these components, and instructions for a robot to build this assembly. The output must 1) resemble the requested object and 2) be reliably assembled by a 6 DoF robot arm with a suction gripper. We then present Blox-Net, a GDfRA system that combines generative vision language models with well-established methods in computer vision, simulation, perturbation analysis, motion planning, and physical robot experimentation to solve a class of GDfRA problems with minimal human supervision. Blox-Net achieved a Top-1 accuracy of 63.5% in the ''recognizability'' of its designed assemblies (eg, resembling giraffe as judged by a VLM). These designs, after automated perturbation redesign, were reliably assembled by a robot, achieving near-perfect success across 10 consecutive assembly iterations with human intervention only during reset prior to assembly. Surprisingly, this entire design process from textual word (''giraffe'') to reliable physical assembly is performed with zero human intervention., Comment: 8 pages, 7 Figures

We explore how to enhance next-token prediction models to perform in-context imitation learning on a real robot, where the robot executes new tasks by interpreting contextual information provided during the input phase, without updating its underlying policy parameters. We propose In-Context Robot Transformer (ICRT), a causal transformer that performs autoregressive prediction on sensorimotor trajectories without relying on any linguistic data or reward function. This formulation enables flexible and training-free execution of new tasks at test time, achieved by prompting the model with sensorimotor trajectories of the new task composing of image observations, actions and states tuples, collected through human teleoperation. Experiments with a Franka Emika robot demonstrate that the ICRT can adapt to new tasks specified by prompts, even in environment configurations that differ from both the prompt and the training data. In a multitask environment setup, ICRT significantly outperforms current state-of-the-art next-token prediction models in robotics on generalizing to unseen tasks. Code, checkpoints and data are available on https://icrt.dev/

Dynamic manipulation of free-end cables has applications for cable management in homes, warehouses and manufacturing plants. We present a supervised learning approach for dynamic manipulation of free-end cables, focusing on the problem of getting the cable endpoint to a designated target position, which may lie outside the reachable workspace of the robot end effector. We present a simulator, tune it to closely match experiments with physical cables, and then collect training data for learning dynamic cable manipulation. We evaluate with 3 cables and a physical UR5 robot. Results over 32x5 trials on 3 cables suggest that a physical UR5 robot can attain a median error distance ranging from 22% to 35% of the cable length among cables, outperforming an analytic baseline by 21% and a Gaussian Process baseline by 7% with lower interquartile range (IQR).

Physics-based simulations have accelerated progress in robot learning for driving, manipulation, and locomotion. Yet, a fast, accurate, and robust surgical simulation environment remains a challenge. In this paper, we present ORBIT-Surgical, a physics-based surgical robot simulation framework with photorealistic rendering in NVIDIA Omniverse. We provide 14 benchmark surgical tasks for the da Vinci Research Kit (dVRK) and Smart Tissue Autonomous Robot (STAR) which represent common subtasks in surgical training. ORBIT-Surgical leverages GPU parallelization to train reinforcement learning and imitation learning algorithms to facilitate study of robot learning to augment human surgical skills. ORBIT-Surgical also facilitates realistic synthetic data generation for active perception tasks. We demonstrate ORBIT-Surgical sim-to-real transfer of learned policies onto a physical dVRK robot. Project website: orbit-surgical.github.io

Inventory monitoring in homes, factories, and retail stores relies on maintaining data despite objects being swapped, added, removed, or moved. We introduce Lifelong LERF, a method that allows a mobile robot with minimal compute to jointly optimize a dense language and geometric representation of its surroundings. Lifelong LERF maintains this representation over time by detecting semantic changes and selectively updating these regions of the environment, avoiding the need to exhaustively remap. Human users can query inventory by providing natural language queries and receiving a 3D heatmap of potential object locations. To manage the computational load, we use Fog-ROS2, a cloud robotics platform, to offload resource-intensive tasks. Lifelong LERF obtains poses from a monocular RGBD SLAM backend, and uses these poses to progressively optimize a Language Embedded Radiance Field (LERF) for semantic monitoring. Experiments with 3-5 objects arranged on a tabletop and a Turtlebot with a RealSense camera suggest that Lifelong LERF can persistently adapt to changes in objects with up to 91% accuracy., Comment: See project webpage at: https://sites.google.com/berkeley.edu/lifelonglerf/home

Touch is an important sensing modality for humans, but it has not yet been incorporated into a multimodal generative language model. This is partially due to the difficulty of obtaining natural language labels for tactile data and the complexity of aligning tactile readings with both visual observations and language descriptions. As a step towards bridging that gap, this work introduces a new dataset of 44K in-the-wild vision-touch pairs, with English language labels annotated by humans (10%) and textual pseudo-labels from GPT-4V (90%). We use this dataset to train a vision-language-aligned tactile encoder for open-vocabulary classification and a touch-vision-language (TVL) model for text generation using the trained encoder. Results suggest that by incorporating touch, the TVL model improves (+29% classification accuracy) touch-vision-language alignment over existing models trained on any pair of those modalities. Although only a small fraction of the dataset is human-labeled, the TVL model demonstrates improved visual-tactile understanding over GPT-4V (+12%) and open-source vision-language models (+32%) on a new touch-vision understanding benchmark. Code and data: https://tactile-vlm.github.io.

With technology for digital photography and high resolution displays rapidly evolving and gaining popularity, there is a growing demand for blind image quality assessment (BIQA) models for high resolution images. Unfortunately, the publicly available large scale image quality databases used for training BIQA models contain mostly low or general resolution images. Since image resizing affects image quality, we assume that the accuracy of BIQA models trained on low resolution images would not be optimal for high resolution images. Therefore, we created a new high resolution image quality database (HRIQ), consisting of 1120 images with resolution of 2880x2160 pixels. We conducted a subjective study to collect the subjective quality ratings for HRIQ in a controlled laboratory setting, resulting in accurate MOS at high resolution. To demonstrate the importance of a high resolution image quality database for training BIQA models to predict mean opinion scores (MOS) of high resolution images accurately, we trained and tested several traditional and deep learning based BIQA methods on different resolution versions of our database. The database is publicly available in https://github.com/jarikorhonen/hriq.

This paper focuses on the problem of detecting and reacting to changes in the distribution of a sensorimotor controller's observables. The key idea is the design of switching policies that can take conformal quantiles as input, which we define as conformal policy learning, that allows robots to detect distribution shifts with formal statistical guarantees. We show how to design such policies by using conformal quantiles to switch between base policies with different characteristics, e.g. safety or speed, or directly augmenting a policy observation with a quantile and training it with reinforcement learning. Theoretically, we show that such policies achieve the formal convergence guarantees in finite time. In addition, we thoroughly evaluate their advantages and limitations on two compelling use cases: simulated autonomous driving and active perception with a physical quadruped. Empirical results demonstrate that our approach outperforms five baselines. It is also the simplest of the baseline strategies besides one ablation. Being easy to use, flexible, and with formal guarantees, our work demonstrates how conformal prediction can be an effective tool for sensorimotor learning under uncertainty., Comment: Conformal Policy Learning

This paper is devoted to the development of a localized Large Language Model (LLM) specifically for Arabic, a language imbued with unique cultural characteristics inadequately addressed by current mainstream models. Significant concerns emerge when addressing cultural sensitivity and local values. To address this, the paper proposes a comprehensive solution that includes further pre-training with Arabic texts, Supervised Fine-Tuning (SFT) utilizing native Arabic instructions, and GPT-4 responses in Arabic, alongside Reinforcement Learning with AI Feedback (RLAIF) employing a reward model attuned to local culture and values. The goal is to cultivate culturally cognizant and value-aligned Arabic LLMs capable of accommodating the diverse, application-specific needs of Arabic-speaking communities. Comprehensive evaluations reveal that the resulting model, dubbed `AceGPT', sets the state-of-the-art standard for open Arabic LLMs across various benchmarks. Codes, data, and models are in https://github.com/FreedomIntelligence/AceGPT., Comment: Accepted to NAACL main conference. https://github.com/FreedomIntelligence/AceGPT

Extensive research has been conducted on assessing grasp stability, a crucial prerequisite for achieving optimal grasping strategies, including the minimum force grasping policy. However, existing works employ basic feature-level fusion techniques to combine visual and tactile modalities, resulting in the inadequate utilization of complementary information and the inability to model interactions between unimodal features. This work proposes an attention-guided cross-modality fusion architecture to comprehensively integrate visual and tactile features. This model mainly comprises convolutional neural networks (CNNs), self-attention, and cross-attention mechanisms. In addition, most existing methods collect datasets from real-world systems, which is time-consuming and high-cost, and the datasets collected are comparatively limited in size. This work establishes a robotic grasping system through physics simulation to collect a multimodal dataset. To address the sim-to-real transfer gap, we propose a migration strategy encompassing domain randomization and domain adaptation techniques. The experimental results demonstrate that the proposed fusion framework achieves markedly enhanced prediction performance (approximately 10%) compared to other baselines. Moreover, our findings suggest that the trained model can be reliably transferred to real robotic systems, indicating its potential to address real-world challenges., Comment: Accepted by IROS 2023

For locomotion, is an arm on a legged robot a liability or an asset for locomotion? Biological systems evolved additional limbs beyond legs that facilitates postural control. This work shows how a manipulator can be an asset for legged locomotion at high speeds or under external perturbations, where the arm serves beyond manipulation. Since the system has 15 degrees of freedom (twelve for the legged robot and three for the arm), off-the-shelf reinforcement learning (RL) algorithms struggle to learn effective locomotion policies. Inspired by Bernstein's neurophysiological theory of animal motor learning, we develop an incremental training procedure that initially freezes some degrees of freedom and gradually releases them, using behaviour cloning (BC) from an early learning procedure to guide optimization in later learning. Simulation experiments show that our policy increases the success rate by up to 61 percentage points over the baselines. Simulation and real robot experiments suggest that our policy learns to use the arm as a tail to initiate robot turning at high speeds and to stabilize the quadruped under external perturbations. Quantitatively, in simulation experiments, we cut the failure rate up to 43.6% during high-speed turning and up to 31.8% for quadruped under external forces compared to using a locked arm.

Moving objects to find a fully-occluded target object, known as mechanical search, is a challenging problem in robotics. As objects are often organized semantically, we conjecture that semantic information about object relationships can facilitate mechanical search and reduce search time. Large pretrained vision and language models (VLMs and LLMs) have shown promise in generalizing to uncommon objects and previously unseen real-world environments. In this work, we propose a novel framework called Semantic Mechanical Search (SMS). SMS conducts scene understanding and generates a semantic occupancy distribution explicitly using LLMs. Compared to methods that rely on visual similarities offered by CLIP embeddings, SMS leverages the deep reasoning capabilities of LLMs. Unlike prior work that uses VLMs and LLMs as end-to-end planners, which may not integrate well with specialized geometric planners, SMS can serve as a plug-in semantic module for downstream manipulation or navigation policies. For mechanical search in closed-world settings such as shelves, we compare with a geometric-based planner and show that SMS improves mechanical search performance by 24% across the pharmacy, kitchen, and office domains in simulation and 47.1% in physical experiments. For open-world real environments, SMS can produce better semantic distributions compared to CLIP-based methods, with the potential to be integrated with downstream navigation policies to improve object navigation tasks. Code, data, videos, and the appendix are available: https://sites.google.com/view/semantic-mechanical-search

Vascular shunt insertion is a fundamental surgical procedure used to temporarily restore blood flow to tissues. It is often performed in the field after major trauma. We formulate a problem of automated vascular shunt insertion and propose a pipeline to perform Automated Vascular Shunt Insertion (AVSI) using a da Vinci Research Kit. The pipeline uses a learned visual model to estimate the locus of the vessel rim, plans a grasp on the rim, and moves to grasp at that point. The first robot gripper then pulls the rim to stretch open the vessel with a dilation motion. The second robot gripper then proceeds to insert a shunt into the vessel phantom (a model of the blood vessel) with a chamfer tilt followed by a screw motion. Results suggest that AVSI achieves a high success rate even with tight tolerances and varying vessel orientations up to 30{\deg}. Supplementary material, dataset, videos, and visualizations can be found at https://sites.google.com/berkeley.edu/autolab-avsi., Comment: Published in: IEEE International Conference on Robotics and Automation (ICRA) 2023

Recently, UAVs endowed with high mobility, low cost, and remote control have promoted the development of UAV-assisted real-time video/image acquisition applications, which have a high demand for both transmission rate and image resolution. However, in conventional vertical photography model, the UAV should fly to the top of ground targets (GTs) to capture images, thus enlarge the transmission delay. In this paper, we propose an oblique photography model, which allows the UAV to capture images of GTs from a far distance while still satisfying the predetermined resolution requirement. Based on the proposed oblique photography model, we further study the UAV placement problem in the cellular-connected UAV-assisted image acquisition system, which aims at minimizing the data transmission delay under the condition of satisfying the predetermined image resolution requirement. Firstly, the proposed scheme is first formulated as an intractable non-convex optimization problem. Then, the original problem is simplified to obtain a tractable suboptimal solution with the help of the block coordinate descent and the successive convex approximation techniques. Finally, the numerical results are presented to show the effectiveness of the proposed scheme. The numerical results have shown that the proposed scheme can largely save the transmission time as compared to the conventional vertical photography model., Comment: submitted to ieee for possible publication. arXiv admin note: text overlap with arXiv:2006.14438 by other authors

Industrial insertion tasks are often performed repetitively with parts that are subject to tight tolerances and prone to breakage. Learning an industrial insertion policy in real is challenging as the collision between the parts and the environment can cause slippage or breakage of the part. In this paper, we present a safe self-supervised method to learn a visuo-tactile insertion policy that is robust to grasp pose variations. The method reduces human input and collisions between the part and the receptacle. The method divides the insertion task into two phases. In the first align phase, a tactile-based grasp pose estimation model is learned to align the insertion part with the receptacle. In the second insert phase, a vision-based policy is learned to guide the part into the receptacle. The robot uses force-torque sensing to achieve a safe self-supervised data collection pipeline. Physical experiments on the USB insertion task from the NIST Assembly Taskboard suggest that the resulting policies can achieve 45/45 insertion successes on 45 different initial grasp poses, improving on two baselines: (1) a behavior cloning agent trained on 50 human insertion demonstrations (1/45) and (2) an online RL policy (TD3) trained in real (0/45).