Direct Preference Optimization Using Sparse Feature-Level Constraints

The alignment of large language models (LLMs) with human preferences remains a key challenge. While post-training techniques like Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) have achieved notable success, they often introduce computational inefficiencies and training instability. In this paper, we propose Feature-level constrained Preference Optimization (FPO), a novel method designed to simplify the alignment process while ensuring stability. FPO leverages pre-trained Sparse Autoencoders (SAEs) and introduces feature-level constraints, allowing for efficient, sparsity-enforced alignment. Our approach enjoys efficiency by using sparse features activated in a well-trained sparse autoencoder and the quality of sequential KL divergence by using the feature-level offline reference. Experimental results on benchmark datasets demonstrate that FPO achieves a 5.08% absolute improvement in win rate with much lower computational cost compared to state-of-the-art baselines, making it a promising solution for efficient and controllable LLM alignments.

Online Omnidirectional Jumping Trajectory Planning for Quadrupedal Robots on Uneven Terrains

Natural terrain complexity often necessitates agile movements like jumping in animals to improve traversal efficiency. To enable similar capabilities in quadruped robots, complex real-time jumping maneuvers are required. Current research does not adequately address the problem of online omnidirectional jumping and neglects the robot's kinodynamic constraints during trajectory generation. This paper proposes a general and complete cascade online optimization framework for omnidirectional jumping for quadruped robots. Our solution systematically encompasses jumping trajectory generation, a trajectory tracking controller, and a landing controller. It also incorporates environmental perception to navigate obstacles that standard locomotion cannot bypass, such as jumping from high platforms. We introduce a novel jumping plane to parameterize omnidirectional jumping motion and formulate a tightly coupled optimization problem accounting for the kinodynamic constraints, simultaneously optimizing CoM trajectory, Ground Reaction Forces (GRFs), and joint states. To meet the online requirements, we propose an accelerated evolutionary algorithm as the trajectory optimizer to address the complexity of kinodynamic constraints. To ensure stability and accuracy in environmental perception post-landing, we introduce a coarse-to-fine relocalization method that combines global Branch and Bound (BnB) search with Maximum a Posteriori (MAP) estimation for precise positioning during navigation and jumping. The proposed framework achieves jump trajectory generation in approximately 0.1 seconds with a warm start and has been successfully validated on two quadruped robots on uneven terrains. Additionally, we extend the framework's versatility to humanoid robots.

CycleResearcher: Improving Automated Research via Automated Review

The automation of scientific discovery has been a long-standing goal within the research community, driven by the potential to accelerate knowledge creation. While significant progress has been made using commercial large language models (LLMs) as research assistants or idea generators, the possibility of automating the entire research process with open-source LLMs remains largely unexplored. This paper explores the feasibility of using open-source post-trained LLMs as autonomous agents capable of performing the full cycle of automated research and review, from literature review and manuscript preparation to peer review and paper revision. Our iterative preference training framework consists of CycleResearcher, which conducts research tasks, and CycleReviewer, which simulates the peer review process, providing iterative feedback via reinforcement learning. To train these models, we develop two new datasets, Review-5k and Research-14k, reflecting real-world machine learning research and peer review dynamics. Our results demonstrate that CycleReviewer achieves a 26.89\% improvement in mean absolute error (MAE) over individual human reviewers in predicting paper scores, indicating that LLMs can surpass expert-level performance in research evaluation. In research, the papers generated by the CycleResearcher model achieved a score of 5.36 in simulated peer reviews, surpassing the preprint level of 5.24 from human experts and approaching the accepted paper level of 5.69. This work represents a significant step toward fully automated scientific inquiry, providing ethical safeguards and advancing AI-driven research capabilities. The code, dataset and model weight are released at \url{http://github/minjun-zhu/Researcher}.

Traversability-Aware Legged Navigation by Learning from Real-World Visual Data

The enhanced mobility brought by legged locomotion empowers quadrupedal robots to navigate through complex and unstructured environments. However, optimizing agile locomotion while accounting for the varying energy costs of traversing different terrains remains an open challenge. Most previous work focuses on planning trajectories with traversability cost estimation based on human-labeled environmental features. However, this human-centric approach is insufficient because it does not account for the varying capabilities of the robot locomotion controllers over challenging terrains. To address this, we develop a novel traversability estimator in a robot-centric manner, based on the value function of the robot's locomotion controller. This estimator is integrated into a new learning-based RGBD navigation framework. The framework develops a planner that guides the robot in avoiding obstacles and hard-to-traverse terrains while reaching its goals. The training of the navigation planner is directly performed in the real world using a sample efficient reinforcement learning method. Through extensive benchmarking, we demonstrate that the proposed framework achieves the best performance in accurate traversability cost estimation and efficient learning from multi-modal data (the robot's color and depth vision, and proprioceptive feedback) for real-world training. Using the proposed method, a quadrupedal robot learns to perform traversability-aware navigation through trial and error in various real-world environments with challenging terrains that are difficult to classify using depth vision alone.

LLM-based MOFs Synthesis Condition Extraction using Few-Shot Demonstrations

The extraction of Metal-Organic Frameworks (MOFs) synthesis conditions from literature text has been challenging but crucial for the logical design of new MOFs with desirable functionality. The recent advent of large language models (LLMs) provides disruptively new solution to this long-standing problem and latest researches have reported over 90% F1 in extracting correct conditions from MOFs literature. We argue in this paper that most existing synthesis extraction practices with LLMs stay with the primitive zero-shot learning, which could lead to downgraded extraction and application performance due to the lack of specialized knowledge. This work pioneers and optimizes the few-shot in-context learning paradigm for LLM extraction of material synthesis conditions. First, we propose a human-AI joint data curation process to secure high-quality ground-truth demonstrations for few-shot learning. Second, we apply a BM25 algorithm based on the retrieval-augmented generation (RAG) technique to adaptively select few-shot demonstrations for each MOF's extraction. Over a dataset randomly sampled from 84,898 well-defined MOFs, the proposed few-shot method achieves much higher average F1 performance (0.93 vs. 0.81, +14.8%) than the native zero-shot LLM using the same GPT-4 model, under fully automatic evaluation that are more objective than the previous human evaluation. The proposed method is further validated through real-world material experiments: compared with the baseline zero-shot LLM, the proposed few-shot approach increases the MOFs structural inference performance (R^2) by 29.4% in average.

Segment, Lift and Fit: Automatic 3D Shape Labeling from 2D Prompts

This paper proposes an algorithm for automatically labeling 3D objects from 2D point or box prompts, especially focusing on applications in autonomous driving. Unlike previous arts, our auto-labeler predicts 3D shapes instead of bounding boxes and does not require training on a specific dataset. We propose a Segment, Lift, and Fit (SLF) paradigm to achieve this goal. Firstly, we segment high-quality instance masks from the prompts using the Segment Anything Model (SAM) and transform the remaining problem into predicting 3D shapes from given 2D masks. Due to the ill-posed nature of this problem, it presents a significant challenge as multiple 3D shapes can project into an identical mask. To tackle this issue, we then lift 2D masks to 3D forms and employ gradient descent to adjust their poses and shapes until the projections fit the masks and the surfaces conform to surrounding LiDAR points. Notably, since we do not train on a specific dataset, the SLF auto-labeler does not overfit to biased annotation patterns in the training set as other methods do. Thus, the generalization ability across different datasets improves. Experimental results on the KITTI dataset demonstrate that the SLF auto-labeler produces high-quality bounding box annotations, achieving an AP@0.5 IoU of nearly 90\%. Detectors trained with the generated pseudo-labels perform nearly as well as those trained with actual ground-truth annotations. Furthermore, the SLF auto-labeler shows promising results in detailed shape predictions, providing a potential alternative for the occupancy annotation of dynamic objects.

Adaptive Model Predictive Control with Data-driven Error Model for Quadrupedal Locomotion

Model Predictive Control (MPC) relies heavily on the robot model for its control law. However, a gap always exists between the reduced-order control model with uncertainties and the real robot, which degrades its performance. To address this issue, we propose the controller of integrating a data-driven error model into traditional MPC for quadruped robots. Our approach leverages real-world data from sensors to compensate for defects in the control model. Specifically, we employ the Autoregressive Moving Average Vector (ARMAV) model to construct the state error model of the quadruped robot using data. The predicted state errors are then used to adjust the predicted future robot states generated by MPC. By such an approach, our proposed controller can provide more accurate inputs to the system, enabling it to achieve desired states even in the presence of model parameter inaccuracies or disturbances. The proposed controller exhibits the capability to partially eliminate the disparity between the model and the real-world robot, thereby enhancing the locomotion performance of quadruped robots. We validate our proposed method through simulations and real-world experimental trials on a large-size quadruped robot that involves carrying a 20 kg un-modeled payload (84% of body weight).

A Fast Online Omnidirectional Quadrupedal Jumping Framework Via Virtual-Model Control and Minimum Jerk Trajectory Generation

Exploring the limits of quadruped robot agility, particularly in the context of rapid and real-time planning and execution of omnidirectional jump trajectories, presents significant challenges due to the complex dynamics involved, especially when considering significant impulse contacts. This paper introduces a new framework to enable fast, omnidirectional jumping capabilities for quadruped robots. Utilizing minimum jerk technology, the proposed framework efficiently generates jump trajectories that exploit its analytical solutions, ensuring numerical stability and dynamic compatibility with minimal computational resources. The virtual model control is employed to formulate a Quadratic Programming (QP) optimization problem to accurately track the Center of Mass (CoM) trajectories during the jump phase. The whole-body control strategies facilitate precise and compliant landing motion. Moreover, the different jumping phase is triggered by time-schedule. The framework's efficacy is demonstrated through its implementation on an enhanced version of the open-source Mini Cheetah robot. Omnidirectional jumps-including forward, backward, and other directional-were successfully executed, showcasing the robot's capability to perform rapid and consecutive jumps with an average trajectory generation and tracking solution time of merely 50 microseconds.

AutoSurvey: Large Language Models Can Automatically Write Surveys

This paper introduces AutoSurvey, a speedy and well-organized methodology for automating the creation of comprehensive literature surveys in rapidly evolving fields like artificial intelligence. Traditional survey paper creation faces challenges due to the vast volume and complexity of information, prompting the need for efficient survey methods. While large language models (LLMs) offer promise in automating this process, challenges such as context window limitations, parametric knowledge constraints, and the lack of evaluation benchmarks remain. AutoSurvey addresses these challenges through a systematic approach that involves initial retrieval and outline generation, subsection drafting by specialized LLMs, integration and refinement, and rigorous evaluation and iteration. Our contributions include a comprehensive solution to the survey problem, a reliable evaluation method, and experimental validation demonstrating AutoSurvey's effectiveness.We open our resources at \url{https://github.com/AutoSurveys/AutoSurvey}.

An Analysis on Quantizing Diffusion Transformers

Diffusion Models (DMs) utilize an iterative denoising process to transform random noise into synthetic data. Initally proposed with a UNet structure, DMs excel at producing images that are virtually indistinguishable with or without conditioned text prompts. Later transformer-only structure is composed with DMs to achieve better performance. Though Latent Diffusion Models (LDMs) reduce the computational requirement by denoising in a latent space, it is extremely expensive to inference images for any operating devices due to the shear volume of parameters and feature sizes. Post Training Quantization (PTQ) offers an immediate remedy for a smaller storage size and more memory-efficient computation during inferencing. Prior works address PTQ of DMs on UNet structures have addressed the challenges in calibrating parameters for both activations and weights via moderate optimization. In this work, we pioneer an efficient PTQ on transformer-only structure without any optimization. By analysing challenges in quantizing activations and weights for diffusion transformers, we propose a single-step sampling calibration on activations and adapt group-wise quantization on weights for low-bit quantization. We demonstrate the efficiency and effectiveness of proposed methods with preliminary experiments on conditional image generation.