Hierarchical Preference Optimization: Learning to achieve goals via feasible subgoals prediction

This work introduces Hierarchical Preference Optimization (HPO), a novel approach to hierarchical reinforcement learning (HRL) that addresses non-stationarity and infeasible subgoal generation issues when solving complex robotic control tasks. HPO leverages maximum entropy reinforcement learning combined with token-level Direct Preference Optimization (DPO), eliminating the need for pre-trained reference policies that are typically unavailable in challenging robotic scenarios. Mathematically, we formulate HRL as a bi-level optimization problem and transform it into a primitive-regularized DPO formulation, ensuring feasible subgoal generation and avoiding degenerate solutions. Extensive experiments on challenging robotic navigation and manipulation tasks demonstrate impressive performance of HPO, where it shows an improvement of up to 35% over the baselines. Furthermore, ablation studies validate our design choices, and quantitative analyses confirm the ability of HPO to mitigate non-stationarity and infeasible subgoal generation issues in HRL.

AIME: AI System Optimization via Multiple LLM Evaluators

Text-based AI system optimization typically involves a feedback loop scheme where a single LLM generates an evaluation in natural language of the current output to improve the next iteration's output. However, in this work, we empirically demonstrate that for a practical and complex task (code generation) with multiple criteria to evaluate, utilizing only one LLM evaluator tends to let errors in generated code go undetected, thus leading to incorrect evaluations and ultimately suboptimal test case performance. Motivated by this failure case, we assume there exists an optimal evaluation policy that samples an evaluation between response and ground truth. We then theoretically prove that a linear combination of multiple evaluators can approximate this optimal policy. From this insight, we propose AI system optimization via Multiple LLM Evaluators (AIME). AIME is an evaluation protocol that utilizes multiple LLMs that each independently generate an evaluation on separate criteria and then combine them via concatenation. We provide an extensive empirical study showing AIME outperforming baseline methods in code generation tasks, with up to $62\%$ higher error detection rate and up to $16\%$ higher success rate than a single LLM evaluation protocol on LeetCodeHard and HumanEval datasets. We also show that the selection of the number of evaluators and which criteria to utilize is non-trivial as it can impact pact success rate by up to $12\%$.

DIPPER: Direct Preference Optimization to Accelerate Primitive-Enabled Hierarchical Reinforcement Learning

Learning control policies to perform complex robotics tasks from human preference data presents significant challenges. On the one hand, the complexity of such tasks typically requires learning policies to perform a variety of subtasks, then combining them to achieve the overall goal. At the same time, comprehensive, well-engineered reward functions are typically unavailable in such problems, while limited human preference data often is; making efficient use of such data to guide learning is therefore essential. Methods for learning to perform complex robotics tasks from human preference data must overcome both these challenges simultaneously. In this work, we introduce DIPPER: Direct Preference Optimization to Accelerate Primitive-Enabled Hierarchical Reinforcement Learning, an efficient hierarchical approach that leverages direct preference optimization to learn a higher-level policy and reinforcement learning to learn a lower-level policy. DIPPER enjoys improved computational efficiency due to its use of direct preference optimization instead of standard preference-based approaches such as reinforcement learning from human feedback, while it also mitigates the well-known hierarchical reinforcement learning issues of non-stationarity and infeasible subgoal generation due to our use of primitive-informed regularization inspired by a novel bi-level optimization formulation of the hierarchical reinforcement learning problem. To validate our approach, we perform extensive experimental analysis on a variety of challenging robotics tasks, demonstrating that DIPPER outperforms hierarchical and non-hierarchical baselines, while ameliorating the non-stationarity and infeasible subgoal generation issues of hierarchical reinforcement learning.

PIPER: Primitive-Informed Preference-based Hierarchical Reinforcement Learning via Hindsight Relabeling

In this work, we introduce PIPER: Primitive-Informed Preference-based Hierarchical reinforcement learning via Hindsight Relabeling, a novel approach that leverages preference-based learning to learn a reward model, and subsequently uses this reward model to relabel higher-level replay buffers. Since this reward is unaffected by lower primitive behavior, our relabeling-based approach is able to mitigate non-stationarity, which is common in existing hierarchical approaches, and demonstrates impressive performance across a range of challenging sparse-reward tasks. Since obtaining human feedback is typically impractical, we propose to replace the human-in-the-loop approach with our primitive-in-the-loop approach, which generates feedback using sparse rewards provided by the environment. Moreover, in order to prevent infeasible subgoal prediction and avoid degenerate solutions, we propose primitive-informed regularization that conditions higher-level policies to generate feasible subgoals for lower-level policies. We perform extensive experiments to show that PIPER mitigates non-stationarity in hierarchical reinforcement learning and achieves greater than 50$\%$ success rates in challenging, sparse-reward robotic environments, where most other baselines fail to achieve any significant progress.

Global Optimality without Mixing Time Oracles in Average-reward RL via Multi-level Actor-Critic

In the context of average-reward reinforcement learning, the requirement for oracle knowledge of the mixing time, a measure of the duration a Markov chain under a fixed policy needs to achieve its stationary distribution-poses a significant challenge for the global convergence of policy gradient methods. This requirement is particularly problematic due to the difficulty and expense of estimating mixing time in environments with large state spaces, leading to the necessity of impractically long trajectories for effective gradient estimation in practical applications. To address this limitation, we consider the Multi-level Actor-Critic (MAC) framework, which incorporates a Multi-level Monte Carlo (MLMC) gradient estimator. With our approach, we effectively alleviate the dependency on mixing time knowledge, a first for average-reward MDPs global convergence. Furthermore, our approach exhibits the tightest-available dependence of $\mathcal{O}\left( \sqrt{\tau_{mix}} \right)$ relative to prior work. With a 2D gridworld goal-reaching navigation experiment, we demonstrate that MAC achieves higher reward than a previous PG-based method for average reward, Parameterized Policy Gradient with Advantage Estimation (PPGAE), especially in cases with relatively small training sample budget restricting trajectory length.

Sampling-based Safe Reinforcement Learning for Nonlinear Dynamical Systems

We develop provably safe and convergent reinforcement learning (RL) algorithms for control of nonlinear dynamical systems, bridging the gap between the hard safety guarantees of control theory and the convergence guarantees of RL theory. Recent advances at the intersection of control and RL follow a two-stage, safety filter approach to enforcing hard safety constraints: model-free RL is used to learn a potentially unsafe controller, whose actions are projected onto safe sets prescribed, for example, by a control barrier function. Though safe, such approaches lose any convergence guarantees enjoyed by the underlying RL methods. In this paper, we develop a single-stage, sampling-based approach to hard constraint satisfaction that learns RL controllers enjoying classical convergence guarantees while satisfying hard safety constraints throughout training and deployment. We validate the efficacy of our approach in simulation, including safe control of a quadcopter in a challenging obstacle avoidance problem, and demonstrate that it outperforms existing benchmarks.

Deceptive Path Planning via Reinforcement Learning with Graph Neural Networks

Deceptive path planning (DPP) is the problem of designing a path that hides its true goal from an outside observer. Existing methods for DPP rely on unrealistic assumptions, such as global state observability and perfect model knowledge, and are typically problem-specific, meaning that even minor changes to a previously solved problem can force expensive computation of an entirely new solution. Given these drawbacks, such methods do not generalize to unseen problem instances, lack scalability to realistic problem sizes, and preclude both on-the-fly tunability of deception levels and real-time adaptivity to changing environments. In this paper, we propose a reinforcement learning (RL)-based scheme for training policies to perform DPP over arbitrary weighted graphs that overcomes these issues. The core of our approach is the introduction of a local perception model for the agent, a new state space representation distilling the key components of the DPP problem, the use of graph neural network-based policies to facilitate generalization and scaling, and the introduction of new deception bonuses that translate the deception objectives of classical methods to the RL setting. Through extensive experimentation we show that, without additional fine-tuning, at test time the resulting policies successfully generalize, scale, enjoy tunable levels of deception, and adapt in real-time to changes in the environment.

Ada-NAV: Adaptive Trajectory-Based Sample Efficient Policy Learning for Robotic Navigation

Reinforcement learning methods, while effective for learning robotic navigation strategies, are known to be highly sample inefficient. This sample inefficiency comes in part from not suitably balancing the explore-exploit dilemma, especially in the presence of non-stationarity, during policy optimization. To incorporate a balance of exploration-exploitation for sample efficiency, we propose Ada-NAV, an adaptive trajectory length scheme where the length grows as a policy's randomness, represented by its Shannon or differential entropy, decreases. Our adaptive trajectory length scheme emphasizes exploration at the beginning of training due to more frequent gradient updates and emphasizes exploitation later on with longer trajectories. In gridworld, simulated robotic environments, and real-world robotic experiments, we demonstrate the merits of the approach over constant and randomly sampled trajectory lengths in terms of performance and sample efficiency. For a fixed sample budget, Ada-NAV results in an 18% increase in navigation success rate, a 20-38% decrease in the navigation path length, and 9.32% decrease in the elevation cost compared to the policies obtained by the other methods. We also demonstrate that Ada-NAV can be transferred and integrated into a Clearpath Husky robot without significant performance degradation.

Beyond Exponentially Fast Mixing in Average-Reward Reinforcement Learning via Multi-Level Monte Carlo Actor-Critic

Many existing reinforcement learning (RL) methods employ stochastic gradient iteration on the back end, whose stability hinges upon a hypothesis that the data-generating process mixes exponentially fast with a rate parameter that appears in the step-size selection. Unfortunately, this assumption is violated for large state spaces or settings with sparse rewards, and the mixing time is unknown, making the step size inoperable. In this work, we propose an RL methodology attuned to the mixing time by employing a multi-level Monte Carlo estimator for the critic, the actor, and the average reward embedded within an actor-critic (AC) algorithm. This method, which we call \textbf{M}ulti-level \textbf{A}ctor-\textbf{C}ritic (MAC), is developed especially for infinite-horizon average-reward settings and neither relies on oracle knowledge of the mixing time in its parameter selection nor assumes its exponential decay; it, therefore, is readily applicable to applications with slower mixing times. Nonetheless, it achieves a convergence rate comparable to the state-of-the-art AC algorithms. We experimentally show that these alleviated restrictions on the technical conditions required for stability translate to superior performance in practice for RL problems with sparse rewards.

Occupancy Information Ratio: Infinite-Horizon, Information-Directed, Parameterized Policy Search

We develop a new measure of the exploration/exploitation trade-off in infinite-horizon reinforcement learning problems called the occupancy information ratio (OIR), which is comprised of a ratio between the infinite-horizon average cost of a policy and the entropy of its long-term state occupancy measure. The OIR ensures that no matter how many trajectories an RL agent traverses or how well it learns to minimize cost, it maintains a healthy skepticism about its environment, in that it defines an optimal policy which induces a high-entropy occupancy measure. Different from earlier information ratio notions, OIR is amenable to direct policy search over parameterized families, and exhibits hidden quasiconcavity through invocation of the perspective transformation. This feature ensures that under appropriate policy parameterizations, the OIR optimization problem has no spurious stationary points, despite the overall problem's nonconvexity. We develop for the first time policy gradient and actor-critic algorithms for OIR optimization based upon a new entropy gradient theorem, and establish both asymptotic and non-asymptotic convergence results with global optimality guarantees. In experiments, these methodologies outperform several deep RL baselines in problems with sparse rewards, where many trajectories may be uninformative and skepticism about the environment is crucial to success.