Reinforcement Learning (RL) has shown promise in improving the reasoning abilities of Large Language Models (LLMs). However, the specific challenges of adapting RL to multimodal data and formats remain relatively unexplored. In this work, we present Observe-R1, a novel framework aimed at enhancing the reasoning capabilities of multimodal large language models (MLLMs). We draw inspirations from human learning progression--from simple to complex and easy to difficult, and propose a gradual learning paradigm for MLLMs. To this end, we construct the NeuraLadder dataset, which is organized and sampled according to the difficulty and complexity of data samples for RL training. To tackle multimodal tasks, we introduce a multimodal format constraint that encourages careful observation of images, resulting in enhanced visual abilities and clearer and more structured responses. Additionally, we implement a bonus reward system that favors concise, correct answers within a length constraint, alongside a dynamic weighting mechanism that prioritizes uncertain and medium-difficulty problems, ensuring that more informative samples have a greater impact on training. Our experiments with the Qwen2.5-VL-3B and Qwen2.5-VL-7B models on 20k samples from the NeuraLadder dataset show that Observe-R1 outperforms a series of larger reasoning models on both reasoning and general benchmarks, achieving superior clarity and conciseness in reasoning chains. Ablation studies validate the effectiveness of our strategies, highlighting the robustness and generalization of our approach. The dataset and code will be released at https://github.com/zrguo/Observe-R1.

The ability to identify and acquire missing information is a critical component of effective decision making and problem solving. With the rise of conversational artificial intelligence (AI) systems, strategically formulating information-seeking questions becomes crucial and demands efficient methods to guide the search process. We introduce a novel approach to adaptive question-asking through a combination of Large Language Models (LLM) for generating questions that maximize information gain, Monte Carlo Tree Search (MCTS) for constructing and leveraging a decision tree across multiple samples, and a hierarchical feedback mechanism to learn from past interactions. We present two key innovations: (1) an adaptive MCTS algorithm that balances exploration and exploitation for efficient search over potential questions; and (2) a clustering-based feedback algorithm that leverages prior experience to guide future interactions. Each incoming sample is assigned to a cluster based on its semantic similarity with previously observed samples. Our UCT (Upper Confidence bound for Trees) formulation selects optimal questions by combining expected rewards, a function of information gain, with a cluster-specific bonus that decays with depth, to emphasize the importance of early-stage questions that have proven effective for narrowing the solution space in similar samples. Experiments across three domains, including medical diagnosis and troubleshooting, demonstrate that our method leads to an average of 12% improvement in success rates and a 10x reduction in the average number of LLM calls made per conversation for the search process, in comparison to the state of the art.

Actor critic methods with sparse rewards in model-based deep reinforcement learning typically require a deterministic binary reward function that reflects only two possible outcomes: if, for each step, the goal has been achieved or not. Our hypothesis is that we can influence an agent to learn faster by applying an external environmental pressure during training, which adversely impacts its ability to get higher rewards. As such, we deviate from the classical paradigm of sparse rewards and add a uniformly sampled reward value to the baseline reward to show that (1) sample efficiency of the training process can be correlated to the adversity experienced during training, (2) it is possible to achieve higher performance in less time and with less resources, (3) we can reduce the performance variability experienced seed over seed, (4) there is a maximum point after which more pressure will not generate better results, and (5) that random positive incentives have an adverse effect when using a negative reward strategy, making an agent under those conditions learn poorly and more slowly. These results have been shown to be valid for Deep Deterministic Policy Gradients using Hindsight Experience Replay in a well known Mujoco environment, but we argue that they could be generalized to other methods and environments as well.

Exploration strategy design is one of the challenging problems in reinforcement learning~(RL), especially when the environment contains a large state space or sparse rewards. During exploration, the agent tries to discover novel areas or high reward~(quality) areas. In most existing methods, the novelty and quality in the neighboring area of the current state are not well utilized to guide the exploration of the agent. To tackle this problem, we propose a novel RL framework, called \underline{c}lustered \underline{r}einforcement \underline{l}earning~(CRL), for efficient exploration in RL. CRL adopts clustering to divide the collected states into several clusters, based on which a bonus reward reflecting both novelty and quality in the neighboring area~(cluster) of the current state is given to the agent. Experiments on a continuous control task and several \emph{Atari 2600} games show that CRL can outperform other state-of-the-art methods to achieve the best performance in most cases.

Count-based exploration algorithms are known to perform near-optimally when used in conjunction with tabular reinforcement learning (RL) methods for solving small discrete Markov decision processes (MDPs). It is generally thought that count-based methods cannot be applied in high-dimensional state spaces, since most states will only occur once. Recent deep RL exploration strategies are able to deal with high-dimensional continuous state spaces through complex heuristics, often relying on optimism in the face of uncertainty or intrinsic motivation. In this work, we describe a surprising finding: a simple generalization of the classic count-based approach can reach near state-of-the-art performance on various high-dimensional and/or continuous deep RL benchmarks. States are mapped to hash codes, which allows to count their occurrences with a hash table. These counts are then used to compute a reward bonus according to the classic count-based exploration theory. We find that simple hash functions can achieve surprisingly good results on many challenging tasks. Furthermore, we show that a domain-dependent learned hash code may further improve these results. Detailed analysis reveals important aspects of a good hash function: 1) having appropriate granularity and 2) encoding information relevant to solving the MDP. This exploration strategy achieves near state-of-the-art performance on both continuous control tasks and Atari 2600 games, hence providing a simple yet powerful baseline for solving MDPs that require considerable exploration.