In this paper we apply model predictive control (MPC), rollout, and reinforcement learning (RL) methodologies to computer chess. We introduce a new architecture for move selection, within which available chess engines are used as components. One engine is used to provide position evaluations in an approximation in value space MPC/RL scheme, while a second engine is used as nominal opponent, to emulate or approximate the moves of the true opponent player. We show that our architecture improves substantially the performance of the position evaluation engine. In other words our architecture provides an additional layer of intelligence, on top of the intelligence of the engines on which it is based. This is true for any engine, regardless of its strength: top engines such as Stockfish and Komodo Dragon (of varying strengths), as well as weaker engines. Structurally, our basic architecture selects moves by a one-move lookahead search, with an intermediate move generated by a nominal opponent engine, and followed by a position evaluation by another chess engine. Simpler schemes that forego the use of the nominal opponent, also perform better than the position evaluator, but not quite by as much. More complex schemes, involving multistep lookahead, may also be used and generally tend to perform better as the length of the lookahead increases. Theoretically, our methodology relies on generic cost improvement properties and the superlinear convergence framework of Newton's method, which fundamentally underlies approximation in value space, and related MPC/RL and rollout/policy iteration schemes. A critical requirement of this framework is that the first lookahead step should be executed exactly. This fact has guided our architectural choices, and is apparently an important factor in improving the performance of even the best available chess engines.

This paper establishes a novel role for Gaussian-mixture models (GMMs) as functional approximators of Q-function losses in reinforcement learning (RL). Unlike the existing RL literature, where GMMs play their typical role as estimates of probability density functions, GMMs approximate here Q-function losses. The new Q-function approximators, coined GMM-QFs, are incorporated in Bellman residuals to promote a Riemannian-optimization task as a novel policy-evaluation step in standard policy-iteration schemes. The paper demonstrates how the hyperparameters (means and covariance matrices) of the Gaussian kernels are learned from the data, opening thus the door of RL to the powerful toolbox of Riemannian optimization. Numerical tests show that with no use of experienced data, the proposed design outperforms state-of-the-art methods, even deep Q-networks which use experienced data, on benchmark RL tasks.

In this paper, we propose a Q-learning based decision-making framework to improve the safety and efficiency of Autonomous Vehicles when they encounter other maliciously behaving vehicles while passing through unsignalized intersections. In Autonomous Vehicles, conventional reward signals are set as regular rewards regarding feedback factors such as safety and efficiency. In this paper, safety gains are modulated by variable weighting parameters to ensure that safety can be emphasized more in emergency situations. The framework proposed in this paper introduces first-order theory of mind inferences on top of conventional rewards, using first-order beliefs as additional reward signals. The decision framework enables Autonomous Vehicles to make informed decisions when encountering vehicles with potentially malicious behaviors at unsignalized intersections, thereby improving the overall safety and efficiency of Autonomous Vehicle transportation systems. In order to verify the performance of the decision framework, this paper uses Prescan/Simulink co-simulations for simulation, and the results show that the performance of the decision framework can meet the set requirements.

While a powerful and promising approach, deep reinforcement learning (DRL) still suffers from sample inefficiency, which can be notably improved by resorting to more sophisticated techniques to address the exploration-exploitation dilemma. One such technique relies on action persistence (i.e., repeating an action over multiple steps). However, previous work exploiting action persistence either applies a fixed strategy or learns additional value functions (or policy) for selecting the repetition number. In this paper, we propose a novel method to dynamically adjust the action persistence based on the current exploration status of the state space. In such a way, our method does not require training of additional value functions or policy. Moreover, the use of a smooth scheduling of the repeat probability allows a more effective balance between exploration and exploitation. Furthermore, our method can be seamlessly integrated into various basic exploration strategies to incorporate temporal persistence. Finally, extensive experiments on different DMControl tasks demonstrate that our state-novelty guided action persistence method significantly improves the sample efficiency.

We consider the problem of estimating the asymptotic variance of a function defined on a Markov chain, an important step for statistical inference of the stationary mean. We design the first recursive estimator that requires $O(1)$ computation at each step, does not require storing any historical samples or any prior knowledge of run-length, and has optimal $O(\frac{1}{n})$ rate of convergence for the mean-squared error (MSE) with provable finite sample guarantees. Here, $n$ refers to the total number of samples generated. The previously best-known rate of convergence in MSE was $O(\frac{\log n}{n})$, achieved by jackknifed estimators, which also do not enjoy these other desirable properties. Our estimator is based on linear stochastic approximation of an equivalent formulation of the asymptotic variance in terms of the solution of the Poisson equation. We generalize our estimator in several directions, including estimating the covariance matrix for vector-valued functions, estimating the stationary variance of a Markov chain, and approximately estimating the asymptotic variance in settings where the state space of the underlying Markov chain is large. We also show applications of our estimator in average reward reinforcement learning (RL), where we work with asymptotic variance as a risk measure to model safety-critical applications. We design a temporal-difference type algorithm tailored for policy evaluation in this context. We consider both the tabular and linear function approximation settings. Our work paves the way for developing actor-critic style algorithms for variance-constrained RL.

Diffusion models have achieved remarkable success in sequential decision-making by leveraging the highly expressive model capabilities in policy learning. A central problem for learning diffusion policies is to align the policy output with human intents in various tasks. To achieve this, previous methods conduct return-conditioned policy generation or Reinforcement Learning (RL)-based policy optimization, while they both rely on pre-defined reward functions. In this work, we propose a novel framework, Forward KL regularized Preference optimization for aligning Diffusion policies, to align the diffusion policy with preferences directly. We first train a diffusion policy from the offline dataset without considering the preference, and then align the policy to the preference data via direct preference optimization. During the alignment phase, we formulate direct preference learning in a diffusion policy, where the forward KL regularization is employed in preference optimization to avoid generating out-of-distribution actions. We conduct extensive experiments for MetaWorld manipulation and D4RL tasks. The results show our method exhibits superior alignment with preferences and outperforms previous state-of-the-art algorithms.

Spatial disorientation is a leading cause of fatal aircraft accidents. This paper explores the potential of AI agents to aid pilots in maintaining balance and preventing unrecoverable losses of control by offering cues and corrective measures that ameliorate spatial disorientation. A multi-axis rotation system (MARS) was used to gather data from human subjects self-balancing in a spaceflight analog condition. We trained models over this data to create "digital twins" that exemplified performance characteristics of humans with different proficiency levels. We then trained various reinforcement learning and deep learning models to offer corrective cues if loss of control is predicted. Digital twins and assistant models then co-performed a virtual inverted pendulum (VIP) programmed with identical physics. From these simulations, we picked the 5 best-performing assistants based on task metrics such as crash frequency and mean distance from the direction of balance. These were used in a co-performance study with 20 new human subjects performing a version of the VIP task with degraded spatial information. We show that certain AI assistants were able to improve human performance and that reinforcement-learning based assistants were objectively more effective but rated as less trusted and preferable by humans.

Intrinsic motivation (IM) and reward shaping are common methods for guiding the exploration of reinforcement learning (RL) agents by adding pseudo-rewards. Designing these rewards is challenging, however, and they can counter-intuitively harm performance. To address this, we characterize them as reward shaping in Bayes-Adaptive Markov Decision Processes (BAMDPs), which formalizes the value of exploration by formulating the RL process as updating a prior over possible MDPs through experience. RL algorithms can be viewed as BAMDP policies; instead of attempting to find optimal algorithms by solving BAMDPs directly, we use it at a theoretical framework for understanding how pseudo-rewards guide suboptimal algorithms. By decomposing BAMDP state value into the value of the information collected plus the prior value of the physical state, we show how psuedo-rewards can help by compensating for RL algorithms' misestimation of these two terms, yielding a new typology of IM and reward shaping approaches. We carefully extend the potential-based shaping theorem to BAMDPs to prove that when pseudo-rewards are BAMDP Potential-based shaping Functions (BAMPFs), they preserve optimal, or approximately optimal, behavior of RL algorithms; otherwise, they can corrupt even optimal learners. We finally give guidance on how to design or convert existing pseudo-rewards to BAMPFs by expressing assumptions about the environment as potential functions on BAMDP states.

Value factorization is a popular paradigm for designing scalable multi-agent reinforcement learning algorithms. However, current factorization methods make choices without full justification that may limit their performance. For example, the theory in prior work uses stateless (i.e., history) functions, while the practical implementations use state information -- making the motivating theory a mismatch for the implementation. Also, methods have built off of previous approaches, inheriting their architectures without exploring other, potentially better ones. To address these concerns, we formally analyze the theory of using the state instead of the history in current methods -- reconnecting theory and practice. We then introduce DuelMIX, a factorization algorithm that learns distinct per-agent utility estimators to improve performance and achieve full expressiveness. Experiments on StarCraft II micromanagement and Box Pushing tasks demonstrate the benefits of our intuitions.

This report describes our proposed solution for the second AI Olympics competition held at IROS 2024. Our solution is based on a recent Model-Based Reinforcement Learning algorithm named MC-PILCO. Besides briefly reviewing the algorithm, we discuss the most critical aspects of the MC-PILCO implementation in the tasks at hand.