Training personal robots using natural language instruction

Talking to Machines: Introducing Robot Perception to Resolve Speech Recognition Uncertainties

“…One aspect that needs to be analysed further is whether, within a particular context (in our case route instructions), the number of different classes from which to discriminate is closed. It has already been shown that although the set of predefined basic primitives required by the robot is not closed, nevertheless there is a low probability (∼2.5%) of new predefined primitives required during an interaction [12]. A similar finding has also been demonstrated in [3] for the lexicon.…”

“…High accuracy in the speech recognition abilities of the system and human-machine interactions based on better evaluative-corrective abilities are therefore the key for the success of this paradigm [13]. Research has been undertaken in the past to define systems able to translate verbal user instructions (or utterances) into machine procedures (see for example [12], [4], [22], [15], [14], [17]). In [22], [15] the verbal interaction allows the robot both to update its current knowledge about its position and to plan the shortest path for a location given by the user.…”

“…In instruction based learning (IBL) [12], a system has been developed to use spontaneous speech to instruct a robot to learn new routes. The system is able to build new sequences of actions (or procedures) starting from an initial set of predefined basic primitives.…”

“…Such models are implemented in Nuance (www.nuance.com), which is used with IBL systems. In [12] the advantages of using Nuance have been discussed. However, some problems cannot be solved with the rule-based approach so far used for Nuance in IBL systems.…”

Spatial language for route-based humanoid robot navigation

“…On the other hand, spatial reasoning concerning verbal route descriptions is essential for humans as well as for mobile robots which navigate in unstructured environments. Such reasoning gives robots the ability to comprehend human-like spatial language, which provides the human users with an intuitive interface that is consistent with their innate spatial cognition (Lauria et al 2001). It can also provide an accelerated learning by using symbolic communication, which was proven by Cangelosi and Harnad 2001. In the last decades, there has been considerable research on spatial language and spatial reasoning.…”