Developing self-directed learners through experiential learning is a transformative approach in education that requires the thoughtful curation of strategies to encourage active engagement, reflection, and application of knowledge. This involves creating real-world learning opportunities and fostering autonomy. We embarked on a study across two groups of students at the National University of Singapore. The first group consists of students from a residence house who participated in an overseas immersion programme in Indonesia. The second group of students from an interest group (NUS Calibur Robotics) participated in an overseas competition, RoboMaster University League in Colorado, Boulder, United States. Our research question in this study was whether experiential learning results in the development of self-directed learning. Several instruments have been adapted, namely the Self-Directed Learning Readiness Scale (SDLRS) originally developed by Murray J. Fisher [1], and revised by Kumar, A.P. [2], as well as the measures of experiential learning and measures of students' approaches to learning and perceived learning generated by Young, M.R. et.al. [3]. Significant improvements were shown across the two groups of students in various dimensions of self-directed learning, like selfmanagement, perceived need for learning, critical thinking and perceived standard of learning. The measurement of all four stages of experiential learning was above 72 %. In summary, this study has demonstrated positive attainment of self-directed learning through experiential learning experiences.

Change detection in videos is a crucial task for identifying alterations in an environment, contingent on knowledge of the base environment. In this work, we assume goal-oriented problems and want to develop an autonomous agent that uses video data from mobile cameras to detect environmental changes. The agent then autonomously adjusts its path in response to these detected changes to reach the target. We introduce an attention-based model by combining UNet and Convolution Blocks Attention Module (CBAM) for change detection, named CUCD, and use a Deep Q-Network (DQN) reinforcement learning model to build a system for autonomous path planning. To examine the effectiveness of the system, we design a pipeline connecting a drone, our change detection model, and a path planning model. The experiments conducted in both simulation and the real world show that changes in the environment can be effectively detected and the navigation path appropriately replanned using our method.

In an effort to improve the human-AI interaction, we explore how large language models (LLMs) can be guided to generate structured problem specifications in the Planning Domain Definition Language (PDDL) and Relational Dynamic Influence Diagram Language (RDDL) to solve planning problems. Our focus in this work is on generating problem specifications based on natural-language instructions, mimicking the interaction between an AI agent (e.g., an assistive robot) and humans. Agents can then be trained on these generated problem specifications using the usual planning/reinforcement learning algorithms. While simple prompting often produces incomplete or syntactically invalid outputs, we find that combining structured prompting, retrieval-augmented generation (RAG), and iterative refinement can significantly improve success rates; however, we also note that the performance remains sensitive to retrieval quality and model variability. To address these challenges, we present two agent-based frameworks. 1)$\mathbf{R}^{3}$AG (RAG, ReAct, Reflexion-Inspired) integrates retrieval with action reasoning and a self-reflective critique step adapted from Reflexion, enabling the model to revise its output by iterating and regenerating the output. 2)$\mathbf{R}^{2} \mathbf{E C}$(RAG, ReAct, Error-Correction) replaces the regeneration loop with a direct localized repair mechanism. An Error Correct Agent identifies and fixes errors in place, retrieving supporting references from a dedicated vector database without discarding the originally generated specification. In PDDL tasks, REC ($\mathbf{R}^{2} \mathbf{E C}$without ReAct) achieves near-perfect PDDL syntactic validity under deterministic decoding. In reinforcement learning tasks,$\mathbf{R}^{3} \mathbf{A G}$achieves 100 % validity in RDDL domains represented within the vector database, while$\mathbf{R}^{2} \mathbf{E C}$generalizes better in domains not present in the vector database, improving success rates from 0 % to 50 %.

Collaborative learning has consistently been a key driver in enhancing educational outcomes and facilitating deeper knowledge understanding. With the increasing adoption of asyn-chronous learning and larger class sizes, many online platforms are emerging to support collaborative learning. However, there remains a noticeable gap in their practical implementation. In this work, we capture the necessary features these platforms should incorporate to maintain the advantages of traditional collaborative learning while navigating the challenges of online environments. We introduce “Klaytab”, a platform that integrates some of these features, aiming to foster effective virtual collabo-ration for peer reviews in writing-focused courses.

We propose a model-based approach to hierarchical reinforcement learning that exploits shared knowledge and selective execution at different levels of abstraction, to efficiently solve large, complex problems. Our framework adopts a new transition dynamics learning algorithm that identifies the common action-feature combinations of the subtasks, and evaluates the subtask execution choices through simulation. The framework is sample efficient, and tolerates uncertain and incomplete problem characterization of the subtasks. We test the framework on common benchmark problems and complex simulated robotic environments. It compares favorably against the state-of-the-art algorithms, and scales well in very large problems.

We propose a new method for transferring a policy from a source task to a target task in model-based reinforcement learning. Our work is motivated by scenarios where a robotic agent operates in similar but challenging environments, such as hospital wards, differentiated by structural arrangements or obstacles, such as furniture. We address problems that require fast responses adapted from incomplete, prior knowledge of the agent in new scenarios. We present an efficient selective exploration strategy that maximally reuses the source task policy. Reuse efficiency is effected through identifying sub-spaces that are different in the target environment, thus limiting the exploration needed in the target task. We empirically show that SEAPoT performs better in terms of jump starts and cumulative average rewards, as compared to existing state-of-the-art policy reuse methods.

The cost of electricity contributes significantly to the operating expense incurred in hosting cloud services. It is necessary to consider this cost while charging the consumers for their service utilization. In this work, we arrive at a metering mechanism for cloud services, in which the price of a cloud service tracks the variable input cost of electricity from a smart grid. The power-aware cloud metering developed here is a dynamic pricing and billing model where tariff for a cloud service is varied in accordance with the input electricity cost. We arrive at a model for power consumption of virtual machines hosted on the cloud infrastructure. This power consumption model is used in calculating the cost of operation of the service. A cloud instance leased by a consumer is billed based on the cost of operation obtained, and its resource utilization. Experimental results validate the approach presented.

Sharing of physical infrastructure using virtualization presents an opportunity to improve the overall resource utilization. It is extremely important for a Software as a Service (SaaS) provider to understand the characteristics of the business application workload in order to size and place the virtual machine (VM) containing the application. A typical business application has a multi-tier architecture and the application workload is often predictable. Using the knowledge of the application architecture and statistical analysis of the workload, one can obtain an appropriate capacity and a good placement strategy for the corresponding VM. In this paper we propose a tool iCirrus-WoP that determines VM capacity and VM collocation possibilities for a given set of application workloads. We perform an empirical analysis of the approach on a set of business application workloads obtained from geographically distributed data centers. The iCirrus-WoP tool determines the fixed reserved capacity and a shared capacity of a VM which it can share with another collocated VM. Based on the workload variation, the tool determines if the VM should be statically allocated or needs a dynamic placement. To determine the collocation possibility, iCirrus-WoP performs a peak utilization analysis of the workloads. The empirical analysis reveals the possibility of collocating applications running in different time-zones. The VM capacity that the tool recommends, show a possibility of improving the overall utilization of the infrastructure by more than 70% if they are appropriately collocated.