SOCIALGYM: A Framework for Benchmarking Social Robot Navigation

• DOI: 10.48550/arxiv.2109.11011
• Publication Date: 2021-01-01
• Authors: Jarrett Holtz; Joydeep Biswas
• Keywords: Benchmark (surveying); Benchmarking; Mobile Robot Navigation; Social robot

Robots moving safely and in a socially compliant manner dynamic human environments is an essential benchmark for long-term robot autonomy. However, it not feasible to learn social navigation behaviors entirely the real world, as learning data-intensive, challenging make safety guarantees during training. Therefore, simulation-based benchmarks that provide abstractions are required. A framework these would need support wide variety of approaches, be extensible broad range scenarios, abstract away perception problem focus on explicitly. While there have been many proposed solutions, including high fidelity 3D simulators grid world approximations, no existing solution satisfies all aforementioned properties evaluating behaviors. In this work, we propose SOCIALGYM, lightweight 2D simulation environment designed with extensibility mind, scenario built SOCIALGYM. Further, present results compare contrast human-engineered model-based approaches suite off-the-shelf Learning from Demonstration (LfD) Reinforcement (RL) applied navigation. These demonstrate data efficiency, task performance, compliance, transfer capabilities each policies evaluated solid grounding future research.