Unlike popular modularized framework, end-to-end autonomous driving seeks to solve the perception, decision and control problems in an integrated way, which can be more adapting new scenarios easier generalize at scale. However, existing approaches are often lack of interpretability, only deal with simple tasks like lane keeping. In this article, we propose interpretable deep reinforcement learning method for driving, is able handle complex urban scenarios. A sequential latent environment...

Motion planning is a core technique for autonomous driving. Nowadays, there still exists lot of challenges in motion driving complicated environments due to: 1) the need both spatial and temporal highly dynamic environments; 2) nonlinear vehicle models non-convex collision avoidance constraints; 3) high computation efficiency real-time implementation. Iterative linear quadratic regulator (ILQR) an algorithm to solve optimal control problem with system very efficiently. However, it can not...

Safety is a critical concern when applying reinforcement learning (RL) to real-world control tasks. However, existing safe RL works either only consider expected safety constraint violations and fail maintain guarantees, or use overly conservative certificate tools borrowed from theory, which sacrifices reward optimization relies on analytic system models. This letter proposes model-free algorithm that achieves near-zero with high rewards. Our key idea jointly learn policy neural barrier...

We focus on learning the zero-constraint-violation safe policy in model-free reinforcement (RL). Existing RL studies mostly use posterior penalty to penalize dangerous actions, which means they must experience danger learn from danger. Therefore, cannot a zero-violation even after convergence. To handle this problem, we leverage safety-oriented energy functions policies and propose set actor-critic (SSAC) algorithm. The function is designed increase rapidly for potentially locating action...

Solid-state fermentation (SSF) and extrusion are effective methods to improve the nutritional sensory quality of rice bran. The effect processing sequence SSF microbial strains on bran was studied. results showed that first followed by increased contents phenolic, flavonoid γ-oryzanol, but color changed brown. caused damage bioactive components antioxidant activity, significantly content arabinoxylans. difference between two sequences may be related process time substrate induction....

Solid-state fermentation (SSF) and extrusion are effective methods to improve the nutritional sensory quality of rice bran. The effect processing sequence SSF microbial strains on bran was studied. results showed that first followed by increased contents phenolic, flavonoid γ-oryzanol, but color changed brown. caused damage bioactive components antioxidant activity, significantly content arabinoxylans. difference between two sequences may be related process time substrate induction.

The Hamilton-Jacobi-Bellman (HJB) equation serves as the necessary and sufficient condition for optimal solution to continuous-time (CT) control problem (OCP). Compared with infinite-horizon HJB equation, solving of finite-horizon (FH) has been a long-standing challenge, because partial time derivative value function is involved an additional unknown term. To address this problem, study first-time bridges link between terminal-time utility function, thus it facilitates use policy iteration...

Safety is a critical concern when applying reinforcement learning (RL) to real-world control problems. A widely used method for ensuring safety learn barrier function with heuristic feasibility labels that come from expert demonstrations [1] or constraint functions [2]. However, their forward invariant sets fall short of the maximum feasible region because inaccurate labels. This paper proposes an algorithm called iteration (FRI) learns generate accurate The core FRI decay (CDF), which comes...

Safe reinforcement learning (RL) that solves constraint-satisfactory policies provides a promising way to the broader safety-critical applications of RL in real-world problems such as robotics. Among all safe approaches, model-based methods reduce training time violations further due their high sample efficiency. However, lacking safety robustness against model uncertainties remains an issue RL, especially safety. In this paper, we propose distributional reachability certificate (DRC) and...

Background: This study investigates the differences in rumen fermentation parameters and microbial communities yaks aged 3 (MarG), 4 (AprG) 5 months (MayG). As crucial livestock on Qinghai-Tibet Plateau, optimizing yak health productivity relies understanding their structure efficiency. These findings provide insights into adaptation to plateau environment support strategies for improving management. Methods: The analyzed (NH3-N, acetate, propionate, isobutyrate, butyrate, valerate)...

Dynamic game arises as a powerful paradigm for multi-robot planning, which safety constraint satisfaction is crucial. Constrained stochastic games are of particular interest, real-world robots need to operate and satisfy constraints under uncertainty. Existing methods solving handle chance using exponential penalties with hand-tuned weights. However, finding suitable penalty weight nontrivial requires trial error. In this letter, we propose the chance-constrained iterative linear-quadratic...

Partially observable Markov decision process (POMDP) is a general framework for making and control under uncertainty. A large class of POMDP algorithms follows two-step approach, in which the first step to estimate belief state, second solve optimal policy taking state as input. The optimality guarantee their combination relies on so-called separation principle. In this paper, we propose new path prove principle infinite horizon problems both discounted cost average cost. We use nominal...

Reinforcement learning (RL) has achieved remarkable success in complex robotic systems (eg. quadruped locomotion). In previous works, the RL-based controller was typically implemented as a single neural network with concatenated observation input. However, corresponding learned policy is highly task-specific. Since all motors are controlled centralized way, out-of-distribution local observations can impact global through coupled policy. contrast, animals and humans control their limbs...