Recent research highlights the potential of multi-modal foundation models in tackling complex decision-making challenges. However, their large parameters make real-world deployment resource-intensive and often impractical for constrained systems. Reinforcement learning (RL) shows promise for task-specific agents but suffers from high sample complexity, limiting practical applications. To address these challenges, we introduce LVLM to Policy (LVLM2P), a novel framework that distills knowledge from large vision-language models (LVLM) into more efficient RL agents. Our approach leverages the LVLM as a teacher, providing instructional actions based on trajectories collected by the RL agent, which helps reduce less meaningful exploration in the early stages of learning, thereby significantly accelerating the agent’s learning progress. Additionally, by leveraging the LVLM to suggest actions directly from visual observations, we eliminate the need for manual textual descriptors of the environment, enhancing applicability across diverse tasks. Experiments show that LVLM2P significantly enhances the sample efficiency of baseline RL algorithms. The code is available at https://github.com/i22024/LVLM2P

In offline reinforcement learning (RL), learning from fixed datasets presents a promising solution for domains where real-time interaction with the environment is expensive or risky. However, designing dense reward signals for offline dataset requires significant human effort and domain expertise. Reinforcement learning with human feedback (RLHF) has emerged as an alternative, but it remains costly due to the human-in-the-loop process, prompting interest in automated reward generation models. To address this, we propose Reward Generation via Large Vision-Language Models (RG-VLM), which leverages the reasoning capabilities of LVLMs to generate rewards from offline data without human involvement. RG-VLM improves generalization in long-horizon tasks and can be seamlessly integrated with the sparse reward signals to enhance task performance, demonstrating its potential as an auxiliary reward signal.

Designing effective reward functions remains a fundamental challenge in reinforcement learning (RL), as it often requires extensive human effort and domain expertise. While RL from human feedback has been successful in aligning agents with human intent, acquiring high-quality feedback is costly and labor-intensive, limiting its scalability. Recent advancements in foundation models present a promising alternative–leveraging AI-generated feedback to reduce reliance on human supervision in reward learning. Building on this paradigm, we introduce ERL-VLM, an enhanced rating-based RL method that effectively learns reward functions from AI feedback. Unlike prior methods that rely on pairwise comparisons, ERL-VLM queries large vision-language models (VLMs) for absolute ratings of individual trajectories, enabling more expressive feedback and improved sample efficiency. Additionally, we propose key enhancements to rating-based RL, addressing instability issues caused by data imbalance and noisy labels. Through extensive experiments across both low-level and high-level control tasks, we demonstrate that ERL-VLM significantly outperforms existing VLM-based reward generation methods. Our results demonstrate the potential of AI feedback for scaling RL with minimal human intervention, paving the way for more autonomous and efficient reward learning.

Offline reinforcement learning (RL) provides a powerful framework for training robotic agents using pre-collected, suboptimal datasets, eliminating the need for costly, time-consuming, and potentially hazardous online interactions. This is particularly useful in safety-critical real-world applications, where online data collection is expensive and impractical. However, existing offline RL algorithms typically require reward labeled data, which introduces an additional bottleneck: reward function design is itself costly, labor-intensive, and requires significant domain expertise. In this paper, we introduce PLARE, a novel approach that leverages large vision-language models (VLMs) to provide guidance signals for agent training. Instead of relying on manually designed reward functions, PLARE queries a VLM for preference labels on pairs of visual trajectory segments based on a language task description. The policy is then trained directly from these preference labels using a supervised contrastive preference learning objective, bypassing the need to learn explicit reward models. Through extensive experiments on robotic manipulation tasks from the MetaWorld, PLARE achieves performance on par with or surpassing existing state-of-the-art VLM-based reward generation methods. Furthermore, we demonstrate the effectiveness of PLARE in real-world manipulation tasks with a physical robot, further validating its practical applicability.

Recent work in offline reinforcement learning (RL) has demonstrated the effectiveness of formulating decision-making as return-conditioned supervised learning. Notably, the decision transformer (DT) architecture has shown promise across various domains. However, despite its initial success, DTs have underperformed on several challenging datasets in goal-conditioned RL. This limitation stems from the inefficiency of return conditioning for guiding policy learning, particularly in unstructured and suboptimal datasets, resulting in DTs failing to effectively learn temporal compositionality. Moreover, this problem might be further exacerbated in long-horizon sparse-reward tasks. To address this challenge, we propose the Predictive Coding for Decision Transformer (PCDT) framework, which leverages generalized future conditioning to enhance DT methods. PCDT utilizes an architecture that extends the DT framework, conditioned on predictive codings, enabling decision-making based on both past and future factors, thereby improving generalization. Through extensive experiments on eight datasets from the AntMaze and FrankaKitchen environments, our proposed method achieves performance on par with or surpassing existing popular value-based and transformer-based methods in offline goal-conditioned RL. Furthermore, we also evaluate our method on a goal-reaching task with a physical robot.

Recent studies reveal that well-performing reinforcement learning (RL) agents in training often lack resilience against adversarial perturbations during deployment. This highlights the importance of building a robust agent before deploying it in the real world. Most prior works focus on developing robust training-based procedures to tackle this problem, including enhancing the robustness of the deep neural network component itself or adversarially training the agent on strong attacks. In this work, we instead study an input transformation-based defense for RL. Specifically, we propose using a variant of vector quantization (VQ) as a transformation for input observations, which is then used to reduce the space of adversarial attacks during testing, resulting in the transformed observations being less affected by attacks. Our method is computationally efficient and seamlessly integrates with adversarial training, further enhancing the robustness of RL agents against adversarial attacks. Through extensive experiments in multiple environments, we demonstrate that using VQ as the input transformation effectively defends against adversarial attacks on the agent’s observations.

This paper considers a network referred to as SoftGroup for accurate and scalable 3D instance segmentation. Existing state-of-the-art methods produce hard semantic predictions followed by grouping instance segmentation results. Unfortunately, errors stemming from hard decisions propagate into the grouping, resulting in poor overlap between predicted instances and ground truth and substantial false positives. To address the abovementioned problems, SoftGroup allows each point to be associated with multiple classes to mitigate the uncertainty stemming from semantic prediction. It also suppresses false positive instances by learning to categorize them as background. Regarding scalability, the existing fast methods require computational time on the order of tens of seconds on large-scale scenes, which is unsatisfactory and far from applicable for real-time. Our finding is that the$k$-Nearest Neighbor ($k$-NN) module, which serves as the prerequisite of grouping, introduces a computational bottleneck. SoftGroup is extended to resolve this computational bottleneck, referred to as SoftGroup++. The proposed SoftGroup++ reduces time complexity with octree$k$-NN and reduces search space with class-aware pyramid scaling and late devoxelization. Experimental results on various indoor and outdoor datasets demonstrate the efficacy and generality of the proposed SoftGroup and SoftGroup++. Their performances surpass the best-performing baseline by a large margin (6%$\sim$16%) in terms of AP$_{50}$. On datasets with large-scale scenes, SoftGroup++ achieves a 6× speed boost on average compared to SoftGroup. Furthermore, SoftGroup can be extended to perform object detection and panoptic segmentation with nontrivial improvements over existing methods.

Existing state-of-the-art 3D instance segmentation methods perform semantic segmentation followed by grouping. The hard predictions are made when performing semantic segmentation such that each point is associated with a single class. However, the errors stemming from hard decision propagate into grouping that results in (1) low overlaps between the predicted instance with the ground truth and (2) substantial false positives. To address the aforementioned problems, this paper proposes a 3D instance segmentation method referred to as SoftGroup by performing bottom-up soft grouping followed by top-down refinement. SoftGroup allows each point to be associated with multiple classes to mitigate the problems stemming from semantic prediction errors and suppresses false positive instances by learning to categorize them as background. Experimental results on different datasets and multiple evaluation metrics demonstrate the efficacy of SoftGroup. Its performance surpasses the strongest prior method by a significant margin of$+6.2\%$on the ScanNet v2 hidden test set and$+6.8\%$on S3DIS Area 5 in terms of$AP_{50}$. Soft-Group is also fast, running at 345ms per scan with a sin-gle Titan X on ScanNet v2 dataset. The source code and trained models for both datasets are available at https://github.com/thangvubk/SoftGroup.git.

Model agnostic meta-learning (MAML) is a popular state-of-the-art meta-learning algorithm that provides good weight initialization of a model given a variety of learning tasks. The model initialized by provided weight can be fine-tuned to an unseen task despite only using a small amount of samples and within a few adaptation steps. MAML is simple and versatile but requires costly learning rate tuning and careful design of the task distribution which affects its scalability and generalization. This paper proposes a more robust MAML based on an adaptive learning scheme and a prioritization task buffer (PTB) referred to as Robust MAML (RMAML) for improving scalability of training process and alleviating the problem of distribution mismatch. RMAML uses gradient-based hyper-parameter optimization to automatically find the optimal learning rate and uses the PTB to gradually adjust training task distribution toward testing task distribution over the course of training. Experimental results on meta reinforcement learning environments demonstrate a substantial performance gain as well as being less sensitive to hyper-parameter choice and robust to distribution mismatch.

A bounding box commonly serves as the proxy for 2D object detection. However, extending this practice to 3D detection raises sensitivity to localization error. This problem is acute on flat objects since small localization error may lead to low overlaps between the prediction and ground truth. To address this problem, this paper proposes Sphere Region Proposal Network (SphereRPN) which detects objects by learning spheres as opposed to bounding boxes. We demonstrate that spherical proposals are more robust to localization error compared to bounding boxes. The proposed SphereRPN is not only accurate but also fast. Experiment results on the standard ScanNet dataset show that the proposed SphereRPN outperforms the previous state-of-the-art methods by a large margin while being $2 \times$ to $7 \times$ faster. The code will be made publicly available.

Developing an agent in reinforcement learning (RL) that is capable of performing complex control tasks directly from high-dimensional observation such as raw pixels is a challenge as efforts still need to be made towards improving sample efficiency and generalization of RL algorithm. This paper considers a learning framework for a Curiosity Contrastive Forward Dynamics Model (CCFDM) to achieve a more sample-efficient RL based directly on raw pixels. CCFDM incorporates a forward dynamics model (FDM) and performs contrastive learning to train its deep convolutional neural network-based image encoder (IE) to extract conducive spatial and temporal information to achieve a more sample efficiency for RL. In addition, during training, CCFDM provides intrinsic rewards, produced based on FDM prediction error, and encourages the curiosity of the RL agent to improve exploration. The diverge and less-repetitive observations provided by both our exploration strategy and data augmentation available in contrastive learning improve not only the sample efficiency but also the generalization . Performance of existing model-free RL methods such as Soft Actor-Critic built on top of CCFDM outperforms prior state-of-the-art pixel-based RL methods on the DeepMind Control Suite benchmark.