We focus on an unloading problem, typical of the logistics sector, modeled as a sequential pick-and-place task. In this type of task, modern machine learning techniques have shown to work better than classic systems since they are more adaptable to stochasticity and better able to cope with large uncertainties. More specifically, supervised and imitation learning have achieved outstanding results in this regard, with the shortcoming of requiring some form of supervision which is not always obtainable for all settings. On the other hand, reinforcement learning (RL) requires much milder form of supervision but still remains impracticable due to its inefficiency. In this paper, we propose and theoretically motivate a novel Unsupervised Reward Shaping algorithm from expert's observations which relaxes the level of supervision required by the agent and works on improving RL performance in our task.

Leveraging learned strategies in unfamiliar scenarios is fundamental to human intelligence. In reinforcement learning, rationally reusing the policies acquired from other tasks or human experts is critical for tackling problems that are difficult to learn from scratch. In this work, we present a framework called Selective Myopic bEhavior Control~(SMEC), which results from the insight that the short-term behaviors of prior policies are sharable across tasks. By evaluating the behaviors of prior policies via a hybrid value function architecture, SMEC adaptively aggregates the sharable short-term behaviors of prior policies and the long-term behaviors of the task policy, leading to coordinated decisions. Empirical results on a collection of manipulation and locomotion tasks demonstrate that SMEC outperforms existing methods, and validate the ability of SMEC to leverage related prior policies.

We investigate safe multi-agent reinforcement learning, where agents seek to collectively maximize an aggregate sum of local objectives while satisfying their own safety constraints. The objective and constraints are described by {\it general utilities}, i.e., nonlinear functions of the long-term state-action occupancy measure, which encompass broader decision-making goals such as risk, exploration, or imitations. The exponential growth of the state-action space size with the number of agents presents challenges for global observability, further exacerbated by the global coupling arising from agents' safety constraints. To tackle this issue, we propose a primal-dual method utilizing shadow reward and $\kappa$-hop neighbor truncation under a form of correlation decay property, where $\kappa$ is the communication radius. In the exact setting, our algorithm converges to a first-order stationary point (FOSP) at the rate of $\mathcal{O}\left(T^{-2/3}\right)$. In the sample-based setting, we demonstrate that, with high probability, our algorithm requires $\widetilde{\mathcal{O}}\left(\epsilon^{-3.5}\right)$ samples to achieve an $\epsilon$-FOSP with an approximation error of $\mathcal{O}(\phi_0^{2\kappa})$, where $\phi_0\in (0,1)$. Finally, we demonstrate the effectiveness of our model through extensive numerical experiments.

We propose a reinforcement learning based method to identify important configurations that connect reactant and product states along chemical reaction paths. By shooting multiple trajectories from these configurations, we can generate an ensemble of configurations that concentrate on the transition path ensemble. This configuration ensemble can be effectively employed in a neural network-based partial differential equation solver to obtain an approximation solution of a restricted Backward Kolmogorov equation, even when the dimension of the problem is very high. The resulting solution, known as the committor function, encodes mechanistic information for the reaction and can in turn be used to evaluate reaction rates.

Pre-trained language models (PLMs) have been widely used to underpin various downstream tasks. However, the adversarial attack task has found that PLMs are vulnerable to small perturbations. Mainstream methods adopt a detached two-stage framework to attack without considering the subsequent influence of substitution at each step. In this paper, we formally model the adversarial attack task on PLMs as a sequential decision-making problem, where the whole attack process is sequential with two decision-making problems, i.e., word finder and word substitution. Considering the attack process can only receive the final state without any direct intermediate signals, we propose to use reinforcement learning to find an appropriate sequential attack path to generate adversaries, named SDM-Attack. Extensive experimental results show that SDM-Attack achieves the highest attack success rate with a comparable modification rate and semantic similarity to attack fine-tuned BERT. Furthermore, our analyses demonstrate the generalization and transferability of SDM-Attack. The code is available at https://github.com/fduxuan/SDM-Attack.

The deployment of robots in uncontrolled environments requires them to operate robustly under previously unseen scenarios, like irregular terrain and wind conditions. Unfortunately, while rigorous safety frameworks from robust optimal control theory scale poorly to high-dimensional nonlinear dynamics, control policies computed by more tractable "deep" methods lack guarantees and tend to exhibit little robustness to uncertain operating conditions. This work introduces a novel approach enabling scalable synthesis of robust safety-preserving controllers for robotic systems with general nonlinear dynamics subject to bounded modeling error by combining game-theoretic safety analysis with adversarial reinforcement learning in simulation. Following a soft actor-critic scheme, a safety-seeking fallback policy is co-trained with an adversarial "disturbance" agent that aims to invoke the worst-case realization of model error and training-to-deployment discrepancy allowed by the designer's uncertainty. While the learned control policy does not intrinsically guarantee safety, it is used to construct a real-time safety filter (or shield) with robust safety guarantees based on forward reachability rollouts. This shield can be used in conjunction with a safety-agnostic control policy, precluding any task-driven actions that could result in loss of safety. We evaluate our learning-based safety approach in a 5D race car simulator, compare the learned safety policy to the numerically obtained optimal solution, and empirically validate the robust safety guarantee of our proposed safety shield against worst-case model discrepancy.

In this paper, we propose the Phy-DRL: a physics-model-regulated deep reinforcement learning framework for safety-critical autonomous systems. The Phy-DRL is unique in three innovations: i) proactive unknown-unknowns training, ii) conjunctive residual control (i.e., integration of data-driven control and physics-model-based control) and safety- \& stability-sensitive reward, and iii) physics-model-based neural network editing, including link editing and activation editing. Thanks to the concurrent designs, the Phy-DRL is able to 1) tolerate unknown-unknowns disturbances, 2) guarantee mathematically provable safety and stability, and 3) strictly comply with physical knowledge pertaining to Bellman equation and reward. The effectiveness of the Phy-DRL is finally validated by an inverted pendulum and a quadruped robot. The experimental results demonstrate that compared with purely data-driven DRL, Phy-DRL features remarkably fewer learning parameters, accelerated training and enlarged reward, while offering enhanced model robustness and safety assurance.

Model-free reinforcement learning algorithms have exhibited great potential in solving single-task sequential decision-making problems with high-dimensional observations and long horizons, but are known to be hard to generalize across tasks. Model-based RL, on the other hand, learns task-agnostic models of the world that naturally enables transfer across different reward functions, but struggles to scale to complex environments due to the compounding error. To get the best of both worlds, we propose a self-supervised reinforcement learning method that enables the transfer of behaviors across tasks with different rewards, while circumventing the challenges of model-based RL. In particular, we show self-supervised pre-training of model-free reinforcement learning with a number of random features as rewards allows implicit modeling of long-horizon environment dynamics. Then, planning techniques like model-predictive control using these implicit models enable fast adaptation to problems with new reward functions. Our method is self-supervised in that it can be trained on offline datasets without reward labels, but can then be quickly deployed on new tasks. We validate that our proposed method enables transfer across tasks on a variety of manipulation and locomotion domains in simulation, opening the door to generalist decision-making agents.

We study the problem of conservative off-policy evaluation (COPE) where given an offline dataset of environment interactions, collected by other agents, we seek to obtain a (tight) lower bound on a policy's performance. This is crucial when deciding whether a given policy satisfies certain minimal performance/safety criteria before it can be deployed in the real world. To this end, we introduce HAMBO, which builds on an uncertainty-aware learned model of the transition dynamics. To form a conservative estimate of the policy's performance, HAMBO hallucinates worst-case trajectories that the policy may take, within the margin of the models' epistemic confidence regions. We prove that the resulting COPE estimates are valid lower bounds, and, under regularity conditions, show their convergence to the true expected return. Finally, we discuss scalable variants of our approach based on Bayesian Neural Networks and empirically demonstrate that they yield reliable and tight lower bounds in various continuous control environments.

Smooth coordination within a swarm robotic system is essential for the effective execution of collective robot missions. Having efficient communication is key to the successful coordination of swarm robots. This paper proposes a new communication-efficient decentralized cooperative reinforcement learning algorithm for coordinating swarm robots. It is made efficient by hierarchically building on the use of local information exchanges. We consider a case study application of maze solving through cooperation among a group of robots, where the time and costs are minimized while avoiding inter-robot collisions and path overlaps during exploration. With a solid theoretical basis, we extensively analyze the algorithm with realistic CORE network simulations and evaluate it against state-of-the-art solutions in terms of maze coverage percentage and efficiency under communication-degraded environments. The results demonstrate significantly higher coverage accuracy and efficiency while reducing costs and overlaps even in high packet loss and low communication range scenarios.