Yirui Zhou

Shaobo Li, Yirui Zhou, Yuan Xu et al.

We present the design and implementation of a RAG-based AI system benchmarking (RAGPerf) framework for characterizing the system behaviors of RAG pipelines. To facilitate detailed profiling and fine-grained performance analysis, RAGPerf decouples the RAG workflow into several modular components - embedding, indexing, retrieval, reranking, and generation. RAGPerf offers the flexibility for users to configure the core parameters of each component and examine their impact on the end-to-end query performance and quality. RAGPerf has a workload generator to model real-world scenarios by supporting diverse datasets (e.g., text, pdf, code, and audio), different retrieval and update ratios, and query distributions. RAGPerf also supports different embedding models, major vector databases such as LanceDB, Milvus, Qdrant, Chroma, and Elasticsearch, as well as different LLMs for content generation. It automates the collection of performance metrics (i.e., end-to-end query throughput, host/GPU memory footprint, and CPU/GPU utilization) and accuracy metrics (i.e., context recall, query accuracy, and factual consistency). We demonstrate the capabilities of RAGPerf through a comprehensive set of experiments and open source its codebase at GitHub. Our evaluation shows that RAGPerf incurs negligible performance overhead.

Wanying Wang, Yichen Zhu, Yirui Zhou et al.

Generative Adversarial Imitation Learning (GAIL) stands as a cornerstone approach in imitation learning. This paper investigates the gradient explosion in two types of GAIL: GAIL with deterministic policy (DE-GAIL) and GAIL with stochastic policy (ST-GAIL). We begin with the observation that the training can be highly unstable for DE-GAIL at the beginning of the training phase and end up divergence. Conversely, the ST-GAIL training trajectory remains consistent, reliably converging. To shed light on these disparities, we provide an explanation from a theoretical perspective. By establishing a probabilistic lower bound for GAIL, we demonstrate that gradient explosion is an inevitable outcome for DE-GAIL due to occasionally large expert-imitator policy disparity, whereas ST-GAIL does not have the issue with it. To substantiate our assertion, we illustrate how modifications in the reward function can mitigate the gradient explosion challenge. Finally, we propose CREDO, a simple yet effective strategy that clips the reward function during the training phase, allowing the GAIL to enjoy high data efficiency and stable trainability.

Yuan Meng, Xiangtong Yao, Haihui Ye et al.

Embodied long-horizon manipulation requires robotic systems to process multimodal inputs-such as vision and natural language-and translate them into executable actions. However, existing learning-based approaches often depend on large, task-specific datasets and struggle to generalize to unseen scenarios. Recent methods have explored using large language models (LLMs) as high-level planners that decompose tasks into subtasks using natural language and guide pretrained low-level controllers. Yet, these approaches assume perfect execution from low-level policies, which is unrealistic in real-world environments with noise or suboptimal behaviors. To overcome this, we fully discard the pretrained low-level policy and instead use the LLM to directly generate executable code plans within a closed-loop framework. Our planner employs chain-of-thought (CoT)-guided few-shot learning with incrementally structured examples to produce robust and generalizable task plans. Complementing this, a reporter evaluates outcomes using RGB-D and delivers structured feedback, enabling recovery from misalignment and replanning under partial observability. This design eliminates per-step inference, reduces computational overhead, and limits error accumulation that was observed in previous methods. Our framework achieves state-of-the-art performance on 30+ diverse seen and unseen long-horizon tasks across LoHoRavens, CALVIN, Franka Kitchen, and cluttered real-world settings.

Yirui Zhou, Yunfei Jin, Xiaowei Liu et al.

Learning from observations (LfO) replicates expert behavior without needing access to the expert's actions, making it more practical than learning from demonstrations (LfD) in many real-world scenarios. However, directly applying the on-policy training scheme in LfO worsens the sample inefficiency problem, while employing the traditional off-policy training scheme in LfO magnifies the instability issue. This paper seeks to develop an efficient and stable solution for the LfO problem. Specifically, we begin by exploring the generalization capabilities of both the reward function and policy in LfO, which provides a theoretical foundation for computation. Building on this, we modify the policy optimization method in generative adversarial imitation from observation (GAIfO) with distributional soft actor-critic (DSAC), and propose the Mimicking Observations through Distributional Update Learning with adequate Exploration (MODULE) algorithm to solve the LfO problem. MODULE incorporates the advantages of (1) high sample efficiency and training robustness enhancement in soft actor-critic (SAC), and (2) training stability in distributional reinforcement learning (RL). Extensive experiments in MuJoCo environments showcase the superior performance of MODULE over current LfO methods.

Liguo Zhou, Yirui Zhou, Huaming Liu et al.

In the rapidly evolving field of autonomous driving systems, the refinement of path planning algorithms is paramount for navigating vehicles through dynamic environments, particularly in complex urban scenarios. Traditional path planning algorithms, which are heavily reliant on static rules and manually defined parameters, often fall short in such contexts, highlighting the need for more adaptive, learning-based approaches. Among these, behavior cloning emerges as a noteworthy strategy for its simplicity and efficiency, especially within the realm of end-to-end path planning. However, behavior cloning faces challenges, such as covariate shift when employing traditional Manhattan distance as the metric. Addressing this, our study introduces the novel concept of Residual Chain Loss. Residual Chain Loss dynamically adjusts the loss calculation process to enhance the temporal dependency and accuracy of predicted path points, significantly improving the model's performance without additional computational overhead. Through testing on the nuScenes dataset, we underscore the method's substantial advancements in addressing covariate shift, facilitating dynamic loss adjustments, and ensuring seamless integration with end-to-end path planning frameworks. Our findings highlight the potential of Residual Chain Loss to revolutionize planning component of autonomous driving systems, marking a significant step forward in the quest for level 5 autonomous driving system.

Yangchun Zhang, Qiang Liu, Weiming Li et al.

Adversarial inverse reinforcement learning (AIRL) stands as a cornerstone approach in imitation learning, yet it faces criticisms from prior studies. In this paper, we rethink AIRL and respond to these criticisms. Criticism 1 lies in Inadequate Policy Imitation. We show that substituting the built-in algorithm with soft actor-critic (SAC) during policy updating (requires multi-iterations) significantly enhances the efficiency of policy imitation. Criticism 2 lies in Limited Performance in Transferable Reward Recovery Despite SAC Integration. While we find that SAC indeed exhibits a significant improvement in policy imitation, it introduces drawbacks to transferable reward recovery. We prove that the SAC algorithm itself is not feasible to disentangle the reward function comprehensively during the AIRL training process, and propose a hybrid framework, PPO-AIRL + SAC, for a satisfactory transfer effect. Criticism 3 lies in Unsatisfactory Proof from the Perspective of Potential Equilibrium. We reanalyze it from an algebraic theory perspective.