KIT's GuideFetch lets robot dogs work in parallel

Karlsruhe Institute of Technology (KIT) researchers have developed GuideFetch, a task coordination framework that enables heterogeneous teams of assistive robot dogs to handle concurrent navigation and object retrieval tasks. The system addresses a key challenge in multi-robot collaboration: a syntactically valid LLM plan might still violate embodiment constraints, meaning the robots could appear to succeed while failing to complete the actual task.

GuideFetch solves this by normalizing robot, skill, and target aliases before execution and validating proposed actions against registered targets and robot capabilities. In a controlled 2×2 study with 360 executions (90 combinations of scene and seed), all 180 online LLM responses validated without fallback or replay. Sequential execution achieved 72/90 successes while parallel execution achieved 71/90, with parallel execution reducing mean makespan by 41.3% in the 56 cases completed by both schedules. The framework demonstrates how role specialization and action overlap can shorten mission completion times in assistive robotics scenarios.