This paper addresses the problem of maintaining safety during training in Reinforcement Learning (RL), such that the safety constraint violations are bounded at any point during learning. In a variety of RL applications the safety of the agent is particularly important, e.g. autonomous platforms or robots that work in proximity of humans. As enforcing safety during training might severely limit the agent's exploration, we propose here a new architecture that handles the trade-off between efficient progress and safety during exploration. As the exploration progresses, we update via Bayesian inference Dirichlet-Categorical models of the transition probabilities of the Markov decision process that describes the environment dynamics. This paper proposes a way to approximate moments of belief about the risk associated to the action selection policy. We construct those approximations, and prove the convergence results. We propose a novel method for leveraging the expectation approximations to derive an approximate bound on the confidence that the risk is below a certain level. This approach can be easily interleaved with RL and we present experimental results to showcase the performance of the overall architecture.

We present a solution of the swing-up and balance task for the pendubot and acrobot for the participation in the AI Olympics competition at IJCAI 2023. Our solution is based on the Soft Actor Crtic (SAC) reinforcement learning (RL) algorithm for training a policy for the swing-up and entering the region of attraction of a linear quadratic regulator(LQR) controller for stabilizing the double pendulum at the top position. Our controller achieves competitive scores in performance and robustness for both, pendubot and acrobot, problem scenarios.

Generative Adversarial Imitation Learning (GAIL) stands as a cornerstone approach in imitation learning. This paper investigates the gradient explosion in two types of GAIL: GAIL with deterministic policy (DE-GAIL) and GAIL with stochastic policy (ST-GAIL). We begin with the observation that the training can be highly unstable for DE-GAIL at the beginning of the training phase and end up divergence. Conversely, the ST-GAIL training trajectory remains consistent, reliably converging. To shed light on these disparities, we provide an explanation from a theoretical perspective. By establishing a probabilistic lower bound for GAIL, we demonstrate that gradient explosion is an inevitable outcome for DE-GAIL due to occasionally large expert-imitator policy disparity, whereas ST-GAIL does not have the issue with it. To substantiate our assertion, we illustrate how modifications in the reward function can mitigate the gradient explosion challenge. Finally, we propose CREDO, a simple yet effective strategy that clips the reward function during the training phase, allowing the GAIL to enjoy high data efficiency and stable trainability.

Standard imitation learning usually assumes that demonstrations are drawn from an optimal policy distribution. However, in real-world scenarios, every human demonstration may exhibit nearly random behavior and collecting high-quality human datasets can be quite costly. This requires imitation learning can learn from imperfect demonstrations to obtain robotic policies that align human intent. Prior work uses confidence scores to extract useful information from imperfect demonstrations, which relies on access to ground truth rewards or active human supervision. In this paper, we propose a dynamics-based method to evaluate the data confidence scores without above efforts. We develop a generalized confidence-based imitation learning framework called Confidence-based Inverse soft-Q Learning (CIQL), which can employ different optimal policy matching methods by simply changing object functions. Experimental results show that our confidence evaluation method can increase the success rate by $40.3\%$ over the original algorithm and $13.5\%$ over the simple noise filtering.

Novelty adaptation is the ability of an intelligent agent to adjust its behavior in response to changes in its environment. This is an important characteristic of intelligent agents, as it allows them to continue to function effectively in novel or unexpected situations, but still stands as a critical challenge for deep reinforcement learning (DRL). To tackle this challenge, we propose a simple yet effective novel method, NAPPING (Novelty Adaptation Principles Learning), that allows trained DRL agents to respond to different classes of novelties in open worlds rapidly. With NAPPING, DRL agents can learn to adjust the trained policy only when necessary. They can quickly generalize to similar novel situations without affecting the part of the trained policy that still works. To demonstrate the efficiency and efficacy of NAPPING, we evaluate our method on four action domains that are different in reward structures and the type of task. The domains are CartPole and MountainCar (classic control), CrossRoad (path-finding), and AngryBirds (physical reasoning). We compare NAPPING with standard online and fine-tuning DRL methods in CartPole, MountainCar and CrossRoad, and state-of-the-art methods in the more complicated AngryBirds domain. Our evaluation results demonstrate that with our proposed method, DRL agents can rapidly and effectively adjust to a wide range of novel situations across all tested domains.

Proximal Policy Optimization (PPO), a popular on-policy deep reinforcement learning method, employs a stochastic policy for exploration. In this paper, we propose a colored noise-based stochastic policy variant of PPO. Previous research highlighted the importance of temporal correlation in action noise for effective exploration in off-policy reinforcement learning. Building on this, we investigate whether correlated noise can also enhance exploration in on-policy methods like PPO. We discovered that correlated noise for action selection improves learning performance and outperforms the currently popular uncorrelated white noise approach in on-policy methods. Unlike off-policy learning, where pink noise was found to be highly effective, we found that a colored noise, intermediate between white and pink, performed best for on-policy learning in PPO. We examined the impact of varying the amount of data collected for each update by modifying the number of parallel simulation environments for data collection and observed that with a larger number of parallel environments, more strongly correlated noise is beneficial. Due to the significant impact and ease of implementation, we recommend switching to correlated noise as the default noise source in PPO.

Robot crowd navigation has been gaining increasing attention and popularity in various practical applications. In existing research, deep reinforcement learning has been applied to robot crowd navigation by training policies in an online mode. However, this inevitably leads to unsafe exploration, and consequently causes low sampling efficiency during pedestrian-robot interaction. To this end, we propose an offline reinforcement learning based robot crowd navigation algorithm by utilizing pre-collected crowd navigation experience. Specifically, this algorithm integrates a spatial-temporal state into implicit Q-Learning to avoid querying out-of-distribution robot actions of the pre-collected experience, while capturing spatial-temporal features from the offline pedestrian-robot interactions. Experimental results demonstrate that the proposed algorithm outperforms the state-of-the-art methods by means of qualitative and quantitative analysis.

Legged locomotion is arguably the most suited and versatile mode to deal with natural or unstructured terrains. Intensive research into dynamic walking and running controllers has recently yielded great advances, both in the optimal control and reinforcement learning (RL) literature. Hopping is a challenging dynamic task involving a flight phase and has the potential to increase the traversability of legged robots. Model based control for hopping typically relies on accurate detection of different jump phases, such as lift-off or touch down, and using different controllers for each phase. In this paper, we present a end-to-end RL based torque controller that learns to implicitly detect the relevant jump phases, removing the need to provide manual heuristics for state detection. We also extend a method for simulation to reality transfer of the learned controller to contact rich dynamic tasks, resulting in successful deployment on the robot after training without parameter tuning.

We present CrystalBox, a novel, model-agnostic, posthoc explainability framework for Deep Reinforcement Learning (DRL) controllers in the large family of input-driven environments which includes computer systems. We combine the natural decomposability of reward functions in input-driven environments with the explanatory power of decomposed returns. We propose an efficient algorithm to generate future-based explanations across both discrete and continuous control environments. Using applications such as adaptive bitrate streaming and congestion control, we demonstrate CrystalBox's capability to generate high-fidelity explanations. We further illustrate its higher utility across three practical use cases: contrastive explanations, network observability, and guided reward design, as opposed to prior explainability techniques that identify salient features.

Temporal difference (TD) learning is often used to update the estimate of the value function which is used by RL agents to extract useful policies. In this paper, we focus on value function estimation in continual reinforcement learning. We propose to decompose the value function into two components which update at different timescales: a permanent value function, which holds general knowledge that persists over time, and a transient value function, which allows quick adaptation to new situations. We establish theoretical results showing that our approach is well suited for continual learning and draw connections to the complementary learning systems (CLS) theory from neuroscience. Empirically, this approach improves performance significantly on both prediction and control problems.