Why Did I Fail? A Causal-Based Method to Find Explanations for Robot Failures

Abstract

Robot failures in human-centered environments are inevitable. Therefore, the ability of robots to explain such failures is paramount for interacting with humans to increase trust and transparency. To achieve this skill, the main challenges addressed in this letter are I) acquiring enough data to learn a cause-effect model of the environment and II) generating causal explanations based on the obtained model. We address I) by learning a causal Bayesian network from simulation data. Concerning II), we propose a novel method that enables robots to generate contrastive explanations upon task failures. The explanation is based on setting the failure state in contrast with the closest state that would have allowed for successful execution. This state is found through breadth-first search and is based on success predictions from the learned causal model. We assessed our method in two different scenarios I) stacking cubes and II) dropping spheres into a container. The obtained causal models reach a sim2real accuracy of 70% and 72%, respectively. We finally show that our novel method scales over multiple tasks and allows real robots to give failure explanations like "the upper cube was stacked too high and too far to the right of the lower cube."

Causal-Based Approaches to Explain and Learn from Self-Extension—A Review

“…To address the problem of explaining the reasons behind functional failure, Diehl and Ramirez-Amaro [47] propose the learning of a Bayesian causal network from simulations. The model is then successfully transferred to real environments.…”

“…One solution is to manually encode the model using a collection of rules, but it is obviously complex to ensure that this set will be complete [25]. The solution can be to build a causal graph [47,66]. For instance, Stocking et al [67] analyse how to distinguish task-relevant and -irrelevant variables exploring causality.…”

“…Brawer et al [68] propose an approach for a robot to learn an explicit model of cause-and-effect by constructing a structural causal model. Diehl and Ramirez-Amaro [47] describe an approach for learning a causal Bayesian network from simulated task executions. The obtained model was used for transferring the acquired knowledge from simulation to several real robots with different embodiments.…”

CAR-DESPOT: Causally-Informed Online POMDP Planning for Robots in Confounded Environments

“…Existing works on causal discovery in robotics aim to discover action-outcome causal relationships, including from human demonstrations in table-setting tasks [15] and from simulated robot actions in block stacking tasks [16]. In the former, authors learn temporal co-occurrences to approximate causal relationships, while in the latter authors learn the structure and parameterisation of a causal Bayesian network.…”

FINO-Net: A Deep Multimodal Sensor Fusion Framework for Manipulation Failure Detection

“…In this article, we present FINO-Net to sense presence of failures by multimodalexteroceptive sensors without integrating any domain/task specific knowledge even though the robot knows which task it is executing. Previously failure detection [10], [11] and identification problems [12] were addressed at symbolic level [13]. In this study, we present FINO-Net as an data driven framework for monitoring.…”