Environment generation is a growing research topic with a rapidly increasing number of applications in robotics. We identify three lines of research in environment generation which, while focusing on different goals, draw upon similar insights and often use related solution techniques: (1) Finding diverse scenarios to test and evaluate robotic systems. (2) Creating large-scale offline datasets of diverse robotics tasks and environments. (3) Open-ended co-evolution of robots and environments to form interactive curricula, which eventually leads to generalizable robots. 

While each track has a different focus and success criteria, they also share common techniques to generate environments, e.g., using generative models trained with human-authored data, integrating robot performance outcomes in the generation process, and using predictive models of robot behaviors for efficient evaluation. The objectives of this workshop are to bring together researchers in the fields of scenario generation, open-ended co-evolution, procedural content generation, and generative modeling to raise awareness about overlapping solution techniques and identify the key challenges in generating environments that enhance the robustness and capabilities of deployed robotic systems. The workshop will include renowned experts from different fields as invited speakers, presentations from researchers submitting papers of original research, focus groups, poster sessions, and a panel session where the invited speakers will present their unique perspectives about the role that environment generation can play in robot science and systems.

Submission open: May 1, 2023 (see the call for papers for submission instructions)

Submission deadline: May 17, 2023 May 25, 2023 (23:59 UTC-12)

Acceptance notification: June 10, 2023 

Workshop: July 10, 2023

(all times are in Korean Standard Time, UTC+9)

• 09:00 - 09:30: Introduction

• 09:30 - 10:00: Talk by Mark Riedl

• 10:00 - 10:30: Talk by Matt Deitke

• 10:30 - 11:00: Coffee Break

• 11:00 - 11:30: Talk by Jeff Clune

• 11:30 - 12:00: Talk by Aleksandra Faust

• 12:00 - 13:30: Lunch

• 13:30 - 14:00: Talk by Deepak Pathak

• 14:00 - 15:00: Panel (Add questions at https://pigeonhole.at/MAJC4E)

• 15:00 - 15:30: Coffee Break

• 15:30 - 16:00: Talk by Tim Rocktäschel

• 16:00 - 17:10: Paper Lightning Talks:

  • 16:00 - 16:10: Li and Varakantham, Generalizable Policy through Diversity: Improving Unsupervised Environment Design

  • 16:10 - 16:20: Zhang and Yang, Generating a Terrain-Robustness Benchmark for Legged Locomotion: A Prototype via Terrain Authoring and Active Learning

  • 16:20 - 16:30: Tio and Varakantham, Zone of Proximal Development with Difficulty-Conditioned Generators

  • 16:30 - 16:40: Ding et al., Semantically Adversarial Scene Generation with Explicit Knowledge Guidance

  • 16:40 - 16:50: Lu et al., Synthetic Experience Replay

  • 16:50 - 17:00: Bhatt et al., Surrogate Assisted Generation of Human-Robot Interaction Scenarios

  • 17:00 - 17:10: Zhang et al., Multi-Robot Coordination and Layout Design for Automated Warehousing