Robotics education increasingly requires learners to connect design choices to measurable task outcomes under realistic engineering constraints. In practice, however, educational robotics laboratories often face limited reusability of build-and-train workflows when robot configurations or task contexts change, while deep reinforcement learning (DRL) further raises the barrier through unstable training dynamics and difficult-to-interpret process traces. This paper presents Xmobot, a configurable mobile robot platform and reusable build-and-train workflow for educational robotics, together with a tool-grounded agentic learning-support system for iterative DRL development. The proposed system retrieves standardized evidence from training logs, assesses run quality through an interpretable fuzzy-logic model, uses a constrained large language model (LLM) to generate structured experiment plans and learner-facing explanations, and applies deterministic constraint validation to ensure traceable and recoverable reward-design iteration. It also provides instructor-facing analytics that aggregate run-level evidence and intervention histories into compact diagnostic views for monitoring and feedback. In an IsaacLab mobile navigation study over a fixed 10-run horizon, the full tool-grounded system achieved zero constraint violations, a valid-plan rate of 1.000, reached the predefined task-success criterion by run 9, and obtained the highest final success rate (0.659), outperforming Manual (0.339), LLM-Chat (0.277), C-noFuzzy (0.370), and C-noValidate (0.612) in this experimental setting. These results suggest that the proposed framework supports more dependable and interpretable DRL iteration while also making robotics education more actionable for learner reflection, instructor monitoring, and formative feedback.

This paper presents a deep reinforcement learning framework for autonomous navigation based on multi-modal belief state representation learned from LiDAR and depth sensors. To address the challenges posed by partial observability and sensor-specific uncertainty, we propose a probabilistic representation module that models belief states as Gaussian distributions over latent environmental features. Sensor-specific encoders extract structured features from raw LiDAR and depth inputs, which are fused using a Q-value-guided weighting scheme derived from the policy critic. A motion-prediction pretraining strategy and a cross-modal coherence loss are introduced to enhance the alignment and reliability of the learned belief states. The resulting representation is integrated into a Soft Actor–Critic (SAC) framework to enable policy-driven decision-making under uncertainty. Extensive experiments in simulated environments demonstrate that the proposed method improves success rate, navigation efficiency, and generalization. Real-world experiments further validate these findings, with the proposed method outperforming a classical navigation baseline by reducing average travel time by 16% and path length by 4%. These results support the use of probabilistic multi-modal belief modeling for autonomous navigation under partial observability.

Metal rolls in a non-ferrous-metal manufacturing workshop manifest the characteristics of symmetry, multiple scales and mutual covering, which poses great challenges for metal roll detection. To solve this problem, firstly, an efficient attention mechanism algorithm named ECLAM (efficient capture location attendant model) is proposed for capturing spatial position features efficiently, to obtain complete location information for metal rolls in a complex environment. ECLAM can improve the ability to extract the spatial features of backbone networks and reduce the influence of the non-critical background. In addition, in order to give feature maps a larger receptive field and improve the weight of location information in multi-scale feature maps, a nonlinear feature fusion module named LFFM (location feature fusion module) is used to fuse two adjacent feature images. Finally, a multi-scale object detection network named L-MSNet (location-based multi-scale object detection network) based on the combination of ECLAM and LFFM is proposed and used to accurately detect multi-scale metal rolls. In the experiments, multi-scale metal roll images are collected from an actual non-ferrous-metal manufacturing workshop. On this basis, a pixel-level image dataset is constructed. Comparative experiments show that, compared with other object detection methods, L-MSNet can detect multi-scale metal rolls more accurately. The average accuracy is improved by 2% to 5%, and the average accuracy of small and medium-sized objects is also significantly improved by 3% to 6%.