Abstract
We present a neurocomputational controller for robotic manipulation based on the recently developed “neural virtual machine” (NVM). The NVM is a purely neural recurrent architecture that emulates a Turing-complete, purely symbolic virtual machine. We program the NVM with a symbolic algorithm that solves blocks-world restacking problems, and execute it in a robotic simulation environment. Our results show that the NVM-based controller can faithfully replicate the execution traces and performance levels of a traditional non-neural program executing the same restacking procedure. Moreover, after programming the NVM, the neurocomputational encodings of symbolic block stacking knowledge can be fine-tuned to further improve performance, by applying reinforcement learning to the underlying neural architecture.
Highlights
Effective manipulation requires tight integration of low-level motor control and high-level reasoning
Our results confirmed that Neural Virtual Machine” (NVM) and reference implementation of the VM (RVM) tick counts are identical (Figure 8A), serving as a sanity check that the NVM was operating correctly
The NVM and RVM were highly correlated on execution time (Figure 8C), the absolute runtime in seconds was substantially higher for the NVM (Figure 8D)
Summary
Effective manipulation requires tight integration of low-level motor control and high-level reasoning. Gupta and Nau (1991) show that a common formalization of optimal blocks-world planning is NPhard, due to a situation they call “deadlock” in which two (or more) blocks cover each other’s goal positions. Another wellknown issue in block stacking and other planning domains is “Sussman’s anomaly,” in which premature resolution of one subgoal may prevent another sub-goal from being achieved without undoing the first (Sussman, 1973). Many sophisticated planners avoid these issues and perform well on block stacking in the average case
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have