The process of creating training data to teach models is currently driven by humans, who manually analyze model weaknesses and plan how to create data that improves a student model. Approaches using LLMs as annotators reduce human effort, but still require humans to interpret feedback from evaluations and control the LLM to produce data the student needs. Automating this labor-intensive process by creating autonomous data generation agents - or teachers - is desirable, but requires environments that can simulate the feedback-driven, iterative, closed loop of data creation. To enable rapid, scalable testing for such agents and their modules, we introduce DataEnvGym, a testbed of teacher environments for data generation agents. DataEnvGym frames data generation as a sequential decision-making task, involving an agent consisting of a data generation policy (which generates a plan for creating training data) and a data generation engine (which transforms the plan into data), inside an environment that provides student feedback. The agent's goal is to improve student performance. Students are iteratively trained and evaluated on generated data, and their feedback (in the form of errors or weak skills) is reported to the agent after each iteration. DataEnvGym includes multiple teacher environment instantiations across 3 levels of structure in the state representation and action space. More structured environments are based on inferred skills and offer more interpretability and curriculum control. We support 4 domains (math, code, VQA, and tool-use) and test multiple students and teachers. Example agents in our teaching environments can iteratively improve students across tasks and settings. Moreover, we show that environments teach different skill levels and test variants of key modules, pointing to future work in improving data generation agents, engines, and feedback mechanisms.

However, there's a slight catch: Each skill tree is tied to two weapon types. That means if you have a specific weapon in mind that you want to beef up with new abilities, you're going to have to pour skill points into whatever skill tree it's tied to before the option becomes available to unlock those techniques. Techniques for the Blast Sling, a slingshot players will recognize from "Zero Dawn" that fires bombs and deals damage to a large area, are available through the survivor skill tree -- another reason to invest your points there first, as Blast Slings are some of the more powerful weapons you gain access to early on.

We introduce CST, an algorithm for constructing skill trees from demonstration trajectories in continuous reinforcement learning domains. CST uses a changepoint detection method to segment each trajectory into a skill chain by detecting a change of appropriate abstraction, or that a segment is too complex to model as a single skill. The skill chains from each trajectory are then merged to form a skill tree. We demonstrate that CST constructs an appropriate skill tree that can be further refined through learning in a challenging continuous domain, and that it can be used to segment demonstration trajectories on a mobile manipulator into chains of skills where each skill is assigned an appropriate abstraction. Papers published at the Neural Information Processing Systems Conference.

The Borderlands video game franchise, which kicked off a decade ago, is one of the most innovative around in terms of gameplay and gear progression. Gamers quickly fell in love with the game's first installment because of all the loot, shooting, and enemies the developers packed into a dynamic world. So the news that "Borderlands 3" was in development was met with much anticipation. Developed by Gearbox and published by 2K, "Borderlands 3," arrived Sept. 13 ($60-up, for Xbox One, Playstation 4 and PC). How does the game deliver and add to the franchise's stellar reputation?

Reinforcement learning combined with deep neural networks has performed remarkably well in many genres of game recently. It surpassed human-level performance in fixed game environments and turn-based two player board games. However, no research has ever shown a result that surpassed human level in modern complex fighting games, to the best of our knowledge. This is due to the inherent difficulties of modern fighting games, including vast action spaces, real-time constraints, and performance generalizations required for various opponents. We overcame these challenges and made 1v1 battle AI agents for the commercial game, "Blade & Soul". The trained agents competed against five professional gamers and achieved 62% of win rate.This paper presents a practical reinforcement learning method including a novel self-play curriculum and data skipping techniques. Through the curriculum, three different styles of agents are created by reward shaping, and are trained against each other for robust performance. Additionally, this paper suggests data skipping techniques which increased data efficiency and facilitated explorations in vast spaces.

Sekiro: Shadows Die Twice solves Dark Souls' biggest problem.Credit: FromSoftware FromSoftware's upcoming video game diverges from its'Soulsborne' series in many ways according to everything we've seen and heard about the game so far. Instead, you'll play a ninja named Sekiro who is an established character in an established story. As with Bloodborne, combat will also see a number of significant changes, with an emphasis on vertical traversal vis-a-vis your Shinobi-arm, a prosthetic that includes a handy grappling hook. But perhaps the most significant change of all is the fact that this is an action-adventure game rather than an RPG. Of course, the Souls games were never heavy on roleplaying, but there was a pretty significant emphasis on RPG progression.

It's been four years since Canadian studio Bioware seemingly closed out its science fiction RPG series Mass Effect with one of the most controversial (or as some put it, "disappointing") endings in video game history. Next March, however, the beloved series is returning, with a brand-new cast and setting, and some interesting new design features. Coming two years after the developer's acclaimed Dragon Age: Inquisition, it's likely Andromeda will draw on a lot of the ideas and systems from that game, as well as from the Mass Effect canon. But what do we actually know about the next title? Mass Effect has often been beautiful, but it's never looked so much like an actual, modern-day sci-fi movie as Andromeda does in the trailers.

We introduce CST, an algorithm for constructing skill trees from demonstration trajectories in continuous reinforcement learning domains. CST uses a changepoint detection method to segment each trajectory into a skill chain by detecting a change of appropriate abstraction, or that a segment is too complex to model as a single skill. The skill chains from each trajectory are then merged to form a skill tree. We demonstrate that CST constructs an appropriate skill tree that can be further refined through learning in a challenging continuous domain, and that it can be used to segment demonstration trajectories on a mobile manipulator into chains of skills where each skill is assigned an appropriate abstraction.