Behavioral experiments on the trust game have shown that trust and trustworthiness are universal among human beings, contradicting the prediction by assuming \emph{Homo economicus} in orthodox Economics. This means some mechanism must be at work that favors their emergence. Most previous explanations however need to resort to some factors based upon imitative learning, a simple version of social learning. Here, we turn to the paradigm of reinforcement learning, where individuals update their strategies by evaluating the long-term return through accumulated experience. Specifically, we investigate the trust game with the Q-learning algorithm, where each participant is associated with two evolving Q-tables that guide one's decision making as trustor and trustee respectively. In the pairwise scenario, we reveal that high levels of trust and trustworthiness emerge when individuals appreciate both their historical experience and returns in the future. Mechanistically, the evolution of the Q-tables shows a crossover that resembles human's psychological changes. We also provide the phase diagram for the game parameters, where the boundary analysis is conducted. These findings are robust when the scenario is extended to a latticed population. Our results thus provide a natural explanation for the emergence of trust and trustworthiness without external factors involved. More importantly, the proposed paradigm shows the potential in deciphering many puzzles in human behaviors.

The reinforcement learning algorithms that focus on how to compute the gradient and choose next actions, are effectively improved the performance of the agents. However, these algorithms are environment-agnostic. This means that the algorithms did not use the knowledge that has been captured by trajectory. This poses that the algorithms should sample many trajectories to train the model. By considering the essence of environment and how much the agent learn from each scenario in that environment, the strategy of the learning procedure can be changed. The strategy retrieves more informative trajectories, so the agent can learn with fewer trajectory sample. We propose Where2Start algorithm that selects the initial state so that the agent has more instability in vicinity of that state. We show that this kind of selection decreases number of trajectories that should be sampled that the agent reach to acceptable reward. Our experiments shows that Where2Start can improve sample efficiency up to 8 times. Also Where2Start can combined with most of state-of-the-art algorithms and improve that robustness and sample efficiency significantly.

Incorporating Autonomous Vehicles (AVs) into existing transportation systems necessitates examining their coexistence with Human-driven Vehicles (HVs) in mixed traffic environments. Central to this coexistence is the AVs' ability to emulate human-like interaction intentions within traffic scenarios. We introduce a novel framework for planning unprotected left-turn trajectories for AVs, designed to mirror human driving behaviors and effectively communicate social intentions. This framework consists of three phases: trajectory generation, evaluation, and selection.In the trajectory generation phase, we utilize real human-driving trajectory data to establish constraints for a predicted trajectory space, creating candidate motion trajectories that reflect intent. The evaluation phase incorporates maximum entropy inverse reinforcement learning (ME-IRL) to gauge human trajectory preferences, considering aspects like traffic efficiency, driving comfort, and interactive safety. During the selection phase, a Boltzmann distribution-based approach is employed to assign rewards and probabilities to the candidate trajectories, promoting human-like decision-making. We validate our framework using an authentic trajectory dataset and conduct a comparative analysis with various baseline methods. Our results, derived from simulator tests and human-in-the-loop driving experiments, affirm our framework's superiority in mimicking human-like driving, expressing intent, and computational efficiency. For additional information of this research, please visit https://shorturl.at/jqu35.

Equipping drones with target search capabilities is highly desirable for applications in disaster rescue and smart warehouse delivery systems. Multiple intelligent drones that can collaborate with each other and maneuver among obstacles show more effectiveness in accomplishing tasks in a shorter amount of time. However, carrying out collaborative target search (CTS) without prior target information is extremely challenging, especially with a visual drone swarm. In this work, we propose a novel data-efficient deep reinforcement learning (DRL) approach called adaptive curriculum embedded multistage learning (ACEMSL) to address these challenges, mainly 3-D sparse reward space exploration with limited visual perception and collaborative behavior requirements. Specifically, we decompose the CTS task into several subtasks including individual obstacle avoidance, target search, and inter-agent collaboration, and progressively train the agents with multistage learning. Meanwhile, an adaptive embedded curriculum (AEC) is designed, where the task difficulty level (TDL) can be adaptively adjusted based on the success rate (SR) achieved in training. ACEMSL allows data-efficient training and individual-team reward allocation for the visual drone swarm. Furthermore, we deploy the trained model over a real visual drone swarm and perform CTS operations without fine-tuning. Extensive simulations and real-world flight tests validate the effectiveness and generalizability of ACEMSL. The project is available at https://github.com/NTU-UAVG/CTS-visual-drone-swarm.git.

The convergence of deterministic policy gradient under the Hadamard parameterization is studied in the tabular setting and the linear convergence of the algorithm is established. To this end, we first show that the error decreases at an $O(\frac{1}{k})$ rate for all the iterations. Based on this result, we further show that the algorithm has a faster local linear convergence rate after $k_0$ iterations, where $k_0$ is a constant that only depends on the MDP problem and the initialization. To show the local linear convergence of the algorithm, we have indeed established the contraction of the sub-optimal probability $b_s^k$ (i.e., the probability of the output policy $\pi^k$ on non-optimal actions) when $k\ge k_0$.

In our rapidly evolving world, effectively managing power grids has become increasingly challenging, primarily due to rising penetration of renewable energy sources and the growing energy demand. While renewable sources like wind and solar power are crucial on our path towards a 100% clean energy future, they introduce considerable uncertainty in power systems, thereby challenging conventional control strategies. Transmission line congestions are often mitigated using redispatch actions, which entail adjusting the power output of various controllable generators in the network. However, these actions are costly and may not fully resolve all issues. Adaptively changing the network using topological actions, such as line switching and bus switching, is an under-utilized yet very cost-effective strategy for network operators facing rapidly shifting energy patterns and contingencies. To navigate the complex and large combinatorial space of all topological actions, we propose a Hierarchical Multi-Agent Reinforcement Learning (MARL) framework in our paper "Multi-Agent Reinforcement Learning for Power Grid Topology Optimization" [1] (a preprint submitted to PSCC 2024).

Hierarchical reinforcement learning (HRL) has the potential to solve complex long horizon tasks using temporal abstraction and increased exploration. However, hierarchical agents are difficult to train due to inherent non-stationarity. We present primitive enabled adaptive relabeling (PEAR), a two-phase approach where we first perform adaptive relabeling on a few expert demonstrations to generate efficient subgoal supervision, and then jointly optimize HRL agents by employing reinforcement learning (RL) and imitation learning (IL). We perform theoretical analysis to $(i)$ bound the sub-optimality of our approach, and $(ii)$ derive a generalized plug-and-play framework for joint optimization using RL and IL. PEAR uses a handful of expert demonstrations and makes minimal limiting assumptions on the task structure. Additionally, it can be easily integrated with typical model free RL algorithms to produce a practical HRL algorithm. We perform experiments on challenging robotic environments and show that PEAR is able to solve tasks that require long term decision making. We empirically show that PEAR exhibits improved performance and sample efficiency over previous hierarchical and non-hierarchical approaches. We also perform real world robotic experiments on complex tasks and demonstrate that PEAR consistently outperforms the baselines.

A key problem in off-policy Reinforcement Learning (RL) is the mismatch, or distribution shift, between the dataset and the distribution over states and actions visited by the learned policy. This problem is exacerbated in the fully offline setting. The main approach to correct this shift has been through importance sampling, which leads to high-variance gradients. Other approaches, such as conservatism or behavior-regularization, regularize the policy at the cost of performance. In this paper, we propose a new approach for stable off-policy Q-Learning. Our method, Projected Off-Policy Q-Learning (POP-QL), is a novel actor-critic algorithm that simultaneously reweights off-policy samples and constrains the policy to prevent divergence and reduce value-approximation error. In our experiments, POP-QL not only shows competitive performance on standard benchmarks, but also out-performs competing methods in tasks where the data-collection policy is significantly sub-optimal.

Reinforcement Learning from Human Feedback (RLHF) has played a crucial role in the success of large models such as ChatGPT. RLHF is a reinforcement learning framework which combines human feedback to improve learning effectiveness and performance. However, obtaining preferences feedback manually is quite expensive in commercial applications. Some statistical commercial indicators are usually more valuable and always ignored in RLHF. There exists a gap between commercial target and model training. In our research, we will attempt to fill this gap with statistical business feedback instead of human feedback, using AB testing which is a well-established statistical method. Reinforcement Learning from Statistical Feedback (RLSF) based on AB testing is proposed. Statistical inference methods are used to obtain preferences for training the reward network, which fine-tunes the pre-trained model in reinforcement learning framework, achieving greater business value. Furthermore, we extend AB testing with double selections at a single time-point to ANT testing with multiple selections at different feedback time points. Moreover, we design numerical experiences to validate the effectiveness of our algorithm framework.

Historically applied exclusively to perfect information games, depth-limited search with value functions has been key to recent advances in AI for imperfect information games. Most prominent approaches with strong theoretical guarantees require subgame decomposition - a process in which a subgame is computed from public information and player beliefs. However, subgame decomposition can itself require non-trivial computations, and its tractability depends on the existence of efficient algorithms for either full enumeration or generation of the histories that form the root of the subgame. Despite this, no formal analysis of the tractability of such computations has been established in prior work, and application domains have often consisted of games, such as poker, for which enumeration is trivial on modern hardware. Applying these ideas to more complex domains requires understanding their cost. In this work, we introduce and analyze the computational aspects and tractability of filtering histories for subgame decomposition. We show that constructing a single history from the root of the subgame is generally intractable, and then provide a necessary and sufficient condition for efficient enumeration. We also introduce a novel Markov Chain Monte Carlo-based generation algorithm for trick-taking card games - a domain where enumeration is often prohibitively expensive. Our experiments demonstrate its improved scalability in the trick-taking card game Oh Hell. These contributions clarify when and how depth-limited search via subgame decomposition can be an effective tool for sequential decision-making in imperfect information settings.