PlanScope: Learning to Plan Within Decision Scope Does Matter

In the context of autonomous driving, learning-based methods have been promising for the development of planning modules. During the training process of planning modules, directly minimizing the discrepancy between expert-driving logs and planning output is widely deployed. In general, driving logs consist of suddenly appearing obstacles or swiftly changing traffic signals, which typically necessitate swift and nuanced adjustments in driving maneuvers. Concurrently, future trajectories of the vehicles exhibit their long-term decisions, such as adhering to a reference lane or circumventing stationary obstacles. Due to the unpredictable influence of future events in driving logs, reasoning bias could be naturally introduced to learning based planning modules, which leads to a possible degradation of driving performance. To address this issue, we identify the decisions and their corresponding time horizons, and characterize a so-called decision scope by retaining decisions within derivable horizons only, to mitigate the effect of irrational behaviors caused by unpredictable events. This framework employs wavelet transformation based log preprocessing with an effective loss computation approach, rendering the planning model only sensitive to valuable decisions at the current state. Since frequency domain characteristics are extracted in conjunction with time domain features by wavelets, decision information across various frequency bands within the corresponding time horizon can be suitably captured. Furthermore, to achieve valuable decision learning, this framework leverages a transformer based decoder that incrementally generates the detailed profiles of future decisions over multiple steps. Our experiments demonstrate that our proposed method outperforms baselines in terms of driving scores with closed-loop evaluations on the nuPlan dataset.

Ada-K Routing: Boosting the Efficiency of MoE-based LLMs

In the era of Large Language Models (LLMs), Mixture-of-Experts (MoE) architectures offer a promising approach to managing computational costs while scaling up model parameters. Conventional MoE-based LLMs typically employ static Top-K routing, which activates a fixed and equal number of experts for each token regardless of their significance within the context. In this paper, we propose a novel Ada-K routing strategy that dynamically adjusts the number of activated experts for each token, thereby improving the balance between computational efficiency and model performance. Specifically, our strategy incorporates learnable and lightweight allocator modules that decide customized expert resource allocation tailored to the contextual needs for each token. These allocators are designed to be fully pluggable, making it broadly applicable across all mainstream MoE-based LLMs. We leverage the Proximal Policy Optimization (PPO) algorithm to facilitate an end-to-end learning process for this non-differentiable decision-making framework. Extensive evaluations on four popular baseline models demonstrate that our Ada-K routing method significantly outperforms conventional Top-K routing. Compared to Top-K, our method achieves over 25% reduction in FLOPs and more than 20% inference speedup while still improving performance across various benchmarks. Moreover, the training of Ada-K is highly efficient. Even for Mixtral-8x22B, a MoE-based LLM with more than 140B parameters, the training time is limited to 8 hours. Detailed analysis shows that harder tasks, middle layers, and content words tend to activate more experts, providing valuable insights for future adaptive MoE system designs. Both the training code and model checkpoints will be publicly available.

Scaling Offline Model-Based RL via Jointly-Optimized World-Action Model Pretraining

A significant aspiration of offline reinforcement learning (RL) is to develop a generalist agent with high capabilities from large and heterogeneous datasets. However, prior approaches that scale offline RL either rely heavily on expert trajectories or struggle to generalize to diverse unseen tasks. Inspired by the excellent generalization of world model in conditional video generation, we explore the potential of image observation-based world model for scaling offline RL and enhancing generalization on novel tasks. In this paper, we introduce JOWA: Jointly-Optimized World-Action model, an offline model-based RL agent pretrained on multiple Atari games to learn general-purpose representation and decision-making ability. Our method jointly optimizes a world-action model through shared transformer backbone, which stabilize temporal difference learning with large models during pretraining. Moreover, we propose an provably efficient and parallelizable planning algorithm to compensate for the Q-value estimation error and thus search out better policies. Experimental results indicate that our largest agent, with 150 million parameters, achieves 78.9% human-level performance on pretrained games using only 10% subsampled offline data, outperforming existing state-of-the-art large-scale offline RL baselines by 31.6% on averange. Furthermore, JOWA scales favorably with model capacity and can sample-efficiently transfer to novel games using only 5k offline fine-tuning data corresponding to about 4 trajectories per game, which demonstrates superior generalization of JOWA. We will release codes at https://github.com/CJReinforce/JOWA.

Multipath Identification and Mitigation with FDA-MIMO Radar

In smart city development, the automatic detection of structures and vehicles within urban or suburban areas via array radar (airborne or vehicle platforms) becomes crucial. However, the inescapable multipath effect adversely affects the radar's capability to detect and track targets. Frequency Diversity Array (FDA)-MIMO radar offers innovative solutions in mitigating multipath due to its frequency flexibility and waveform diversity traits amongst array elements. Hence, utilizing FDA-MIMO radar, this research proposes a multipath discrimination and suppression strategy to augment target detection and suppress false alarms. The primary advancement is the transformation of conventional multipath suppression into a multipath recognition issue, thereby enabling multipath components from single-frame echo data to be separated without prior knowledge. By offsetting the distance steering vectors of different objects to be detected, the accurate spectral information corresponding to the current distance unit can be extracted during spatial spectrum estimation. The direct and multipath components are differentiated depending on whether the transmitting and receiving angles match. Additionally, to mitigate high-order multipath, the echo intensity of multipath components is reduced via joint optimization of array transmit weighting and frequency increment. The numerical results show that the proposed algorithm can identify multipath at different distances in both single-target and multi-target scenarios, which is superior to the general MIMO radar.

Toward Motion Robustness: A masked attention regularization framework in remote photoplethysmography

There has been growing interest in facial video-based remote photoplethysmography (rPPG) measurement recently, with a focus on assessing various vital signs such as heart rate and heart rate variability. Despite previous efforts on static datasets, their approaches have been hindered by inaccurate region of interest (ROI) localization and motion issues, and have shown limited generalization in real-world scenarios. To address these challenges, we propose a novel masked attention regularization (MAR-rPPG) framework that mitigates the impact of ROI localization and complex motion artifacts. Specifically, our approach first integrates a masked attention regularization mechanism into the rPPG field to capture the visual semantic consistency of facial clips, while it also employs a masking technique to prevent the model from overfitting on inaccurate ROIs and subsequently degrading its performance. Furthermore, we propose an enhanced rPPG expert aggregation (EREA) network as the backbone to obtain rPPG signals and attention maps simultaneously. Our EREA network is capable of discriminating divergent attentions from different facial areas and retaining the consistency of spatiotemporal attention maps. For motion robustness, a simple open source detector MediaPipe for data preprocessing is sufficient for our framework due to its superior capability of rPPG signal extraction and attention regularization. Exhaustive experiments on three benchmark datasets (UBFC-rPPG, PURE, and MMPD) substantiate the superiority of our proposed method, outperforming recent state-of-the-art works by a considerable margin.

RiskMap: A Unified Driving Context Representation for Autonomous Motion Planning in Urban Driving Environment

Planning is complicated by the combination of perception and map information, particularly when driving in heavy traffic. Developing an extendable and efficient representation that visualizes sensor noise and provides constraints to real-time planning tasks is desirable. We aim to develop an extendable map representation offering prior to cost in planning tasks to simplify the planning process of dealing with complex driving scenarios and visualize sensor noise. In this paper, we illustrate a unified context representation empowered by a modern deep learning motion prediction model, representing statistical cognition of motion prediction for human beings. A sampling-based planner is adopted to train and compare the difference in risk map generation methods. The training tools and model structures are investigated illustrating their efficiency in this task.

PLUTO: Pushing the Limit of Imitation Learning-based Planning for Autonomous Driving

We present PLUTO, a powerful framework that pushes the limit of imitation learning-based planning for autonomous driving. Our improvements stem from three pivotal aspects: a longitudinal-lateral aware model architecture that enables flexible and diverse driving behaviors; An innovative auxiliary loss computation method that is broadly applicable and efficient for batch-wise calculation; A novel training framework that leverages contrastive learning, augmented by a suite of new data augmentations to regulate driving behaviors and facilitate the understanding of underlying interactions. We assessed our framework using the large-scale real-world nuPlan dataset and its associated standardized planning benchmark. Impressively, PLUTO achieves state-of-the-art closed-loop performance, beating other competing learning-based methods and surpassing the current top-performed rule-based planner for the first time. Results and code are available at https://jchengai.github.io/pluto.

SC-Tune: Unleashing Self-Consistent Referential Comprehension in Large Vision Language Models

Recent trends in Large Vision Language Models (LVLMs) research have been increasingly focusing on advancing beyond general image understanding towards more nuanced, object-level referential comprehension. In this paper, we present and delve into the self-consistency capability of LVLMs, a crucial aspect that reflects the models' ability to both generate informative captions for specific objects and subsequently utilize these captions to accurately re-identify the objects in a closed-loop process. This capability significantly mirrors the precision and reliability of fine-grained visual-language understanding. Our findings reveal that the self-consistency level of existing LVLMs falls short of expectations, posing limitations on their practical applicability and potential. To address this gap, we introduce a novel fine-tuning paradigm named Self-Consistency Tuning (SC-Tune). It features the synergistic learning of a cyclic describer-locator system. This paradigm is not only data-efficient but also exhibits generalizability across multiple LVLMs. Through extensive experiments, we demonstrate that SC-Tune significantly elevates performance across a spectrum of object-level vision-language benchmarks and maintains competitive or improved performance on image-level vision-language benchmarks. Both our model and code will be publicly available at https://github.com/ivattyue/SC-Tune.

RIME: Robust Preference-based Reinforcement Learning with Noisy Preferences

Preference-based Reinforcement Learning (PbRL) avoids the need for reward engineering by harnessing human preferences as the reward signal. However, current PbRL algorithms over-reliance on high-quality feedback from domain experts, which results in a lack of robustness. In this paper, we present RIME, a robust PbRL algorithm for effective reward learning from noisy preferences. Our method incorporates a sample selection-based discriminator to dynamically filter denoised preferences for robust training. To mitigate the accumulated error caused by incorrect selection, we propose to warm start the reward model, which additionally bridges the performance gap during transition from pre-training to online training in PbRL. Our experiments on robotic manipulation and locomotion tasks demonstrate that RIME significantly enhances the robustness of the current state-of-the-art PbRL method. Ablation studies further demonstrate that the warm start is crucial for both robustness and feedback-efficiency in limited-feedback cases.

Efficient Global Navigational Planning in 3D Structures based on Point Cloud Tomography

Navigation in complex 3D scenarios requires appropriate environment representation for efficient scene understanding and trajectory generation. We propose a highly efficient and extensible global navigation framework based on a tomographic understanding of the environment to navigate ground robots in multi-layer structures. Our approach generates tomogram slices using the point cloud map to encode the geometric structure as ground and ceiling elevations. Then it evaluates the scene traversability considering the robot's motion capabilities. Both the tomogram construction and the scene evaluation are accelerated through parallel computation. Our approach further alleviates the trajectory generation complexity compared with planning in 3D spaces directly. It generates 3D trajectories by searching through multiple tomogram slices and separately adjusts the robot height to avoid overhangs. We evaluate our framework in various simulation scenarios and further test it in the real world on a quadrupedal robot. Our approach reduces the scene evaluation time by 3 orders of magnitude and improves the path planning speed by 3 times compared with existing approaches, demonstrating highly efficient global navigation in various complex 3D environments. The code is available at: https://github.com/byangw/PCT_planner.