iGibson 2.0: Object-Centric Simulation for Robot Learning of Everyday Household Tasks

Recent research in embodied AI has been boosted by the use of simula1 tion environments to develop and train robot learning approaches. However, the 2 use of simulation has skewed the attention to tasks that only require what robotics 3 simulators can simulate: motion and physical contact. We present iGibson 2.0, an 4 open-source simulation environment that supports the simulation of a more diverse 5 set of household tasks through three key innovations. First, iGibson 2.0 supports 6 object states, including temperature, wetness level, cleanliness level, and toggled 7 and sliced states, necessary to cover a wider range of tasks. Second, iGibson 2.0 8 implements a set of predicate logic functions that map the simulator states to logic 9 states like Cooked or Soaked. Additionally, given a logic state, iGibson 2.0 can 10 sample valid physical states that satisfy it. This functionality can generate poten11 tially infinite instances of tasks with minimal effort from the users. The sampling 12 mechanism allows our scenes to be more densely populated with small objects in 13 semantically meaningful locations. Third, iGibson 2.0 includes a virtual reality 14 (VR) interface to immerse humans in its scenes to collect demonstrations. As a 15 result, we can collect demonstrations from humans on these new types of tasks, and 16 use them for imitation learning. We evaluate the new capabilities of iGibson 2.0 to 17 enable robot learning of novel tasks, in the hope of demonstrating the potential of 18 this new simulator to support new research in embodied AI. iGibson 2.0 and its 19 new dataset will be publicly available after the anonymous reviewing process. 20