Reinforcement learning (RL) can help agents learn complex tasks that would be hard to specify using standard imperative programming. However, end users may have trouble personalizing their technology using RL due to a lack of technical expertise. Prior work has explored means of supporting end users after a problem for the RL agent to solve has been defined. Little work, however, has explored how to support end users when defining this problem. We propose a tool to provide structured support for end users defining problems for RL agents. Through this tool, users can (i) directly and indirectly specify the problem as a Markov decision process (MDP); (ii) receive automatic suggestions on possible MDP changes that would enhance training time and accuracy; and (iii) revise the MDP after training the agent to solve it. We believe this work will help reduce barriers to using RL and contribute to the existing literature on designing human-in-the-loop systems.
more »
« less
Mixed-Initiative Level Design with RL Brush
This paper introduces RL Brush, a level-editing tool for tile-based games designed for mixed-initiative co-creation. The tool uses reinforcement-learning-based models to augment manual human level-design through the addition of AI-generated suggestions. Here, we apply RL Brush to designing levels for the classic puzzle game Sokoban. We put the tool online and tested it in 39 different sessions. The results show that users using the AI suggestions stay around longer and their created levels on average are more playable and more complex than without.
more »
« less
- Award ID(s):
- 1717324
- NSF-PAR ID:
- 10231866
- Date Published:
- Journal Name:
- Artificial Intelligence in Music, Sound, Art and Design - 10th International Conference, EvoMUSART 2021, Held as Part of EvoStar 2021
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)Abstract Flooding in coastal cities is increasing due to climate change and sea-level rise, stressing the traditional stormwater systems these communities rely on. Automated real-time control (RTC) of these systems can improve performance, and creating control policies for smart stormwater systems is an active area of study. This research explores reinforcement learning (RL) to create control policies to mitigate flood risk. RL is trained using a model of hypothetical urban catchments with a tidal boundary and two retention ponds with controllable valves. RL's performance is compared to the passive system, a model predictive control (MPC) strategy, and a rule-based control strategy (RBC). RL learns to proactively manage pond levels using current and forecast conditions and reduced flooding by 32% over the passive system. Compared to the MPC approach using a physics-based model and genetic algorithm, RL achieved nearly the same flood reduction, just 3% less than MPC, with a significant 88× speedup in runtime. Compared to RBC, RL was able to quickly learn similar control strategies and reduced flooding by an additional 19%. This research demonstrates that RL can effectively control a simple system and offers a computationally efficient method that could scale to RTC of more complex stormwater systems.more » « less
-
Rodrigo, M.M. ; Matsuda, N. ; Cristea, A.I. ; Dimitrova, V. (Ed.)It might be highly effective if students could transition dynamically between individual and collaborative learning activities, but how could teachers manage such complex classroom scenarios? Although recent work in AIED has focused on teacher tools, little is known about how to orchestrate dynamic transitions between individual and collaborative learning. We created a novel technology ecosystem that supports these dynamic transitions. The ecosystem integrates a novel teacher orchestration tool that provides monitoring support and pairing suggestions with two AI-based tutoring systems that support individual and collaborative learning, respectively. We tested the feasibility of this ecosystem in a classroom study with 5 teachers and 199 students over 22 class sessions. We found that the teachers were able to manage the dynamic transitions and valued them. The study contributes a new technology ecosystem for dynamically transitioning between individual and collaborative learning, plus insight into the orchestration functionality that makes these transitions feasible.more » « less
-
The rapid pace of recent research in AI has been driven in part by the presence of fast and challenging simulation environments. These environments often take the form of games; with tasks ranging from simple board games, to competitive video games. We propose a new benchmark - Obstacle Tower: a high fidelity, 3D, 3rd person, procedurally generated environment. An agent in Obstacle Tower must learn to solve both low-level control and high-level planning problems in tandem while learning from pixels and a sparse reward signal.Unlike other benchmarks such as the Arcade Learning Environment, evaluation of agent performance in Obstacle Tower is based on an agent's ability to perform well on unseen instances of the environment. In this paper we outline the environment and provide a set of baseline results produced by current state-of-the-art Deep RL methods as well as human players. These algorithms fail to produce agents capable of performing near human level.more » « less
-
In this proposal, we will share some initial findings about how teacher and student engagement in cogenerative dialogues influenced the development of the Culturally Relevant Pedagogical Guidelines for Computational Thinking and Computer Science (CRPG-CSCT). The CRPG-CSCT’s purpose is to provide computer science teachers with tools to enhance their instruction by accurately reflecting students’ diverse cultural resources in the classroom. Additionally, the CRPG-CSCT will provide guidance to non-computer science teachers on how to facilitate the integration of computational thinking skills to a broad spectrum of classes in the arts, humanities, sciences, social sciences, and mathematics. Our initial findings shared here are part of a larger NSF-funded research project (Award No. 2122367) which aims to better understand the barriers to entry and challenges for success faced by underrepresented secondary school students in computer science, through direct engagement with the students themselves. Throughout the 2022-23 academic year, the researchers have been working with a small team of secondary school teachers, students, and instructional designers, as well as university faculty in computer science, secondary education, and sociology to develop the CRPG-CSCT. The CRPG-CSCT is rooted in the tenets of culturally relevant pedagogy (Ladson-Billings, 1995) and borrows from Muhammad’s (2020) work in Cultivating Genius: An Equity Framework for Culturally and Historically Responsive Literacy. The CRPG-CCT is being developed over six day-long workshops held throughout the academic year. At the time of this submission, five of the six workshops had been completed. Each workshop utilized cogenerative dialogues (cogens) as the primary tool for organizing and sustaining participants’ engagement. Through cogens, participants more deeply learn about students’ cultural capital and the value of utilizing that capital within the classroom (Roth, Lawless, & Tobin, 2000). The success of cogens relies on following specific protocols (Emdin, 2016), such as listening attentively, ensuring there are equal opportunities for all participants to share, and affirming the experiences of other participants. The goal of a cogen is to reach a collective decision, based on the dialogue, that will positively impact students by explicitly addressing barriers to their engagement in the classroom. During each workshop, one member of the research team and one undergraduate research assistant observed the interactions among cogen participants and documented these in the form of ethnographic field notes. Another undergraduate research assistant took detailed notes during the workshop to record the content of small and large group discussions, presentations, and questions/responses throughout the workshops. A grounded theory approach was used to analyze the field notes. Additionally, at the conclusion of each workshop, participants completed a Cogen Feedback Survey (CFS) to gather additional information. The CFS were analyzed through open thematic coding, memos, and code frequencies. Our preliminary results demonstrate high levels of engagement from teacher and student participants during the workshops. Students identified that the cogen structure allowed them to participate comfortably, openly, and honestly. Further, students described feeling valued and heard. Students’ ideas and experiences were frequently affirmed, which served as an important step toward dismantling traditional teacher-student boundaries that might otherwise prevent them from sharing freely. Another result from the use of cogens was the shared experience of participants comprehending views from the other group’s perspective in the classroom. Students appreciated the opportunity to learn from teachers about their struggles in keeping students engaged. Teachers appreciated the opportunity to better understand students’ schooling experiences and how these may affirm or deny aspects of their identity. Finally, all participants shared meaningful suggestions and strategies for future workshops and for the collective betterment of the group. Initial findings shared here are important for several reasons. First, our findings suggest that cogens are an effective approach for fostering participants’ commitment to creating the conditions for students’ success in the classroom. Within the context of the workshops, cogens provided teachers, students, and faculty with opportunities to engage in authentic conversations for addressing the recruitment and retention problems in computer science for underrepresented students. These conversations often resulted in the development of tangible pedagogical approaches, examples, metaphors, and other strategies to directly address the recruitment and retention of underrepresented students in computer science. Finally, while we are still developing the CRPG-CSCT, cogens provided us with the opportunity to ensure the voices of teachers and students are well represented in and central to the document.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1007/978-3-030-72914-1_27
