Abstract Autonomous navigation between narrow vineyard rows requires respecting the platform’s dynamics, which trajectory optimizers such as CHOMP and STOMP resolve in batch before motion and again whenever the map changes. We instead cast local navigation as a continuous-control Markov decision process whose actions are the per-track torques of a skid-steer robot, so the platform’s dynamics enter the control law itself, and the optimization is paid once, during training. Soft Actor-Critic (SAC) and Twin Delayed Deep Deterministic policy gradient (TD3), both recurrent, and a feed-forward Proximal Policy Optimization (PPO) baseline are trained over ten seeds in simulation from a real vineyard passability map. On two held-out scenarios all three produce shorter routes than a conservative weighted A* reference at higher peak but comparable average impassability, and the safety ranking of the off-policy agents reverses between scenarios.
Filip Zúbek, Oliver Halaš, V. Skokan et al.· Journal of Electrical Engine...· 0 citations
This paper presents a unified technical survey of Large Language Models (LLMs), connecting three layers of the modelling pipeline that existing surveys address in isolation: internal architecture, distributed training optimisation, and downstream adaptation. Its organising principle is the dependency between these layers—how a choice at one constrains what remains feasible at the next. The survey examines fundamental mechanisms (tokenisation, scaled dot-product attention, activation functions, and normalisation, including RMSNorm and pre- versus post-normalisation placement), then the engineering of training at scale: data, tensor, and pipeline parallelism, hybrid schemes, mixed-precision training with BF16 and FP8, ZeRO-Offload memory management, activation checkpointing, and compute-optimal scaling laws together with the conditions under which they fail. The adaptation section covers supervised and instruction fine-tuning, a comparison of parameter-efficient methods (LoRA, QLoRA, adapters, prefix and prompt tuning), Reinforcement Learning from Human Feedback with its reward-hacking failure mode, alternatives including DPO, KTO and Constitutional AI, Retrieval-Augmented Generation beyond the basic pipeline, and decoding strategies. Practical configuration guidance is given for 7B, 70B and trillion-parameter regimes. Dedicated treatments of Mixture-of-Experts architectures, long-context modelling, and hardware-aware co-design close the survey, with open challenges classified by origin and severity.
Martin Lukáč, F. Duchoň, Jakub Ivan et al.· Applied Sciences· 0 citations