There has been a growing interest in developing learner models to enhance learning and teaching experiences in educational environments. However, existing works have primarily focused on structured environments relying on meticulously crafted representations of tasks, thereby limiting the agent's ability to generalize skills across tasks. In this paper, we aim to enhance the generalization capabilities of agents in open-ended text-based learning environments by integrating Reinforcement Learning (RL) with Large Language Models (LLMs). We investigate three types of agents: (i) RL-based agents that utilize natural language for state and action representations to find the best interaction strategy, (ii) LLM-based agents that leverage the model's general knowledge and reasoning through prompting, and (iii) hybrid LLM-assisted RL agents that combine these two strategies to improve agents' performance and generalization. To support the development and evaluation of these agents, we introduce PharmaSimText, a novel benchmark derived from the PharmaSim virtual pharmacy environment designed for practicing diagnostic conversations. Our results show that RL-based agents excel in task completion but lack in asking quality diagnostic questions. In contrast, LLM-based agents perform better in asking diagnostic questions but fall short of completing the task. Finally, hybrid LLM-assisted RL agents enable us to overcome these limitations, highlighting the potential of combining RL and LLMs to develop high-performing agents for open-ended learning environments.

While language models (LMs) offer significant capability in zero-shot reasoning tasks across a wide range of domains, they do not perform satisfactorily in problems which requires multi-step reasoning. Previous approaches to mitigate this involves breaking a larger, multi-step task into sub-tasks and asking the language model to generate proposals ("thoughts") for each sub-task and using exhaustive planning approaches such as DFS to compose a solution. In this work, we leverage this idea to introduce two new contributions: first, we formalize a planning-based approach to perform multi-step problem solving with LMs via Partially Observable Markov Decision Processes (POMDPs), with the LM's own reflections about the value of a state used as a search heuristic; second, leveraging the online POMDP solver POMCP, we demonstrate a superior success rate of 89.4% on the Game of 24 task as compared to existing approaches while also offering better anytime performance characteristics than fixed tree-search which is used previously. Taken together, these contributions allow modern LMs to decompose and solve larger-scale reasoning tasks more effectively.

In the classical Reinforcement Learning from Human Feedback (RLHF) framework, Proximal Policy Optimization (PPO) is employed to learn from sparse, sentence-level rewards -- a challenging scenario in traditional deep reinforcement learning. Despite the great successes of PPO in the alignment of state-of-the-art closed-source large language models (LLMs), its open-source implementation is still largely sub-optimal, as widely reported by numerous research studies. To address these issues, we introduce a framework that models RLHF problems as a Markov decision process (MDP), enabling the capture of fine-grained token-wise information. Furthermore, we provide theoretical insights that demonstrate the superiority of our MDP framework over the previous sentence-level bandit formulation. Under this framework, we introduce an algorithm, dubbed as Reinforced Token Optimization (\texttt{RTO}), which learns the token-wise reward function from preference data and performs policy optimization based on this learned token-wise reward signal. Theoretically, \texttt{RTO} is proven to have the capability of finding the near-optimal policy sample-efficiently. For its practical implementation, \texttt{RTO} innovatively integrates Direct Preference Optimization (DPO) and PPO. DPO, originally derived from sparse sentence rewards, surprisingly provides us with a token-wise characterization of response quality, which is seamlessly incorporated into our subsequent PPO training stage. Extensive real-world alignment experiments verify the effectiveness of the proposed approach.

In recent years, the widespread application of multi-robot systems in areas such as power inspection, autonomous vehicle fleets has made multi-robot technology a research hotspot in the field of robotics. This paper investigates multi-robot cooperative exploration in unknown environments, proposing a training framework and decision strategy based on multi-agent reinforcement learning. Specifically we propose a Asymmetric Topological Representation based mapping framework (ATR-Mapping), combining the advantages of methods based on raw grid maps and methods based on topology, the structural information from the raw grid maps is extracted and combined with a topological graph constructed based on geometric distance information for decision-making. Leveraging this topological graph representation, we employs a decision network based on topological graph matching to assign corresponding boundary points to each robot as long-term target points for decision-making. We conducts testing and application of the proposed algorithms in real world scenarios using the Gazebo and Gibson simulation environments. It validates that the proposed method, when compared to existing methods, achieves a certain degree of performance improvement.

We consider the problem of an autonomous agent equipped with multiple sensors, each with different sensing precision and energy costs. The agent's goal is to explore the environment and gather information subject to its resource constraints in unknown, partially observable environments. The challenge lies in reasoning about the effects of sensing and movement while respecting the agent's resource and dynamic constraints. We formulate the problem as a trajectory optimization problem and solve it using a projection-based trajectory optimization approach where the objective is to reduce the variance of the Gaussian process world belief. Our approach outperforms previous approaches in long horizon trajectories by achieving an overall variance reduction of up to 85% and reducing the root-mean square error in the environment belief by 50%. This approach was developed in support of rover path planning for the NASA VIPER Mission.

We consider the problem of olfactory searches in a turbulent environment. We focus on agents that respond solely to odor stimuli, with no access to spatial perception nor prior information about the odor location. We ask whether navigation strategies to a target can be learned robustly within a sequential decision making framework. We develop a reinforcement learning algorithm using a small set of interpretable olfactory states and train it with realistic turbulent odor cues. By introducing a temporal memory, we demonstrate that two salient features of odor traces, discretized in few olfactory states, are sufficient to learn navigation in a realistic odor plume. Performance is dictated by the sparse nature of turbulent plumes. An optimal memory exists which ignores blanks within the plume and activates a recovery strategy outside the plume. We obtain the best performance by letting agents learn their recovery strategy and show that it is mostly casting cross wind, similar to behavior observed in flying insects. The optimal strategy is robust to substantial changes in the odor plumes, suggesting minor parameter tuning may be sufficient to adapt to different environments.

This paper introduces a decentralized state-dependent Markov chain synthesis (DSMC) algorithm for finite-state Markov chains. We present a state-dependent consensus protocol that achieves exponential convergence under mild technical conditions, without relying on any connectivity assumptions regarding the dynamic network topology. Utilizing the proposed consensus protocol, we develop the DSMC algorithm, updating the Markov matrix based on the current state while ensuring the convergence conditions of the consensus protocol. This result establishes the desired steady-state distribution for the resulting Markov chain, ensuring exponential convergence from all initial distributions while adhering to transition constraints and minimizing state transitions. The DSMC's performance is demonstrated through a probabilistic swarm guidance example, which interprets the spatial distribution of a swarm comprising a large number of mobile agents as a probability distribution and utilizes the Markov chain to compute transition probabilities between states. Simulation results demonstrate faster convergence for the DSMC based algorithm when compared to the previous Markov chain based swarm guidance algorithms.

Manipulating unseen articulated objects through visual feedback is a critical but challenging task for real robots. Existing learning-based solutions mainly focus on visual affordance learning or other pre-trained visual models to guide manipulation policies, which face challenges for novel instances in real-world scenarios. In this paper, we propose a novel part-guided 3D RL framework, which can learn to manipulate articulated objects without demonstrations. We combine the strengths of 2D segmentation and 3D RL to improve the efficiency of RL policy training. To improve the stability of the policy on real robots, we design a Frame-consistent Uncertainty-aware Sampling (FUS) strategy to get a condensed and hierarchical 3D representation. In addition, a single versatile RL policy can be trained on multiple articulated object manipulation tasks simultaneously in simulation and shows great generalizability to novel categories and instances. Experimental results demonstrate the effectiveness of our framework in both simulation and real-world settings. Our code is available at https://github.com/THU-VCLab/Part-Guided-3D-RL-for-Sim2Real-Articulated-Object-Manipulation.

Machines that can replicate human intelligence with type 2 reasoning capabilities should be able to reason at multiple levels of spatio-temporal abstractions and scales using internal world models. Devising formalisms to develop such internal world models, which accurately reflect the causal hierarchies inherent in the dynamics of the real world, is a critical research challenge in the domains of artificial intelligence and machine learning. This thesis identifies several limitations with the prevalent use of state space models (SSMs) as internal world models and propose two new probabilistic formalisms namely Hidden-Parameter SSMs and Multi-Time Scale SSMs to address these drawbacks. The structure of graphical models in both formalisms facilitates scalable exact probabilistic inference using belief propagation, as well as end-to-end learning via backpropagation through time. This approach permits the development of scalable, adaptive hierarchical world models capable of representing nonstationary dynamics across multiple temporal abstractions and scales. Moreover, these probabilistic formalisms integrate the concept of uncertainty in world states, thus improving the system's capacity to emulate the stochastic nature of the real world and quantify the confidence in its predictions. The thesis also discuss how these formalisms are in line with related neuroscience literature on Bayesian brain hypothesis and predicitive processing. Our experiments on various real and simulated robots demonstrate that our formalisms can match and in many cases exceed the performance of contemporary transformer variants in making long-range future predictions. We conclude the thesis by reflecting on the limitations of our current models and suggesting directions for future research.

We would like to enable a collaborative multiagent team to navigate at long length scales and under uncertainty in real-world environments. In practice, planning complexity scales with the number of agents in the team, with the length scale of the environment, and with environmental uncertainty. Enabling tractable planning requires developing abstract models that can represent complex, high-quality plans. However, such models often abstract away information needed to generate directly-executable plans for real-world agents in real-world environments, as planning in such detail, especially in the presence of real-world uncertainty, would be computationally intractable. In this paper, we describe the deployment of a planning system that used a hierarchy of planners to execute collaborative multiagent navigation tasks in real-world, unknown environments. By developing a planning system that was robust to failures at every level of the planning hierarchy, we enabled the team to complete collaborative navigation tasks, even in the presence of imperfect planning abstractions and real-world uncertainty. We deployed our approach on a Clearpath Husky-Jackal team navigating in a structured outdoor environment, and demonstrated that the system enabled the agents to successfully execute collaborative plans.