Adaptive Neuro-Symbolic Planning for bio-inspired soft robotics maintenance in hybrid quantum-classical pipelines
说明与用途
神经符号系统通过将连续传感器数据映射为概率性谓词,传递给符号规划器,并可使用量子退火进行组合动作选择,来规划软体机器人的维护。
问题
Soft robotic actuators degrade gradually, producing continuous degradation signals that classical symbolic planners cannot model; maintenance planning under uncertainty is the bottleneck.
结果
The neuro‑symbolic architecture enabled maintenance planning that explicitly handled uncertainty, allowed policy adjustment via threshold tuning, and leveraged quantum annealing for large actuator sets.
工作流
- Collect continuous sensor data from actuators
- Feed sensor windows into a neural network (PredicateGrounder) to obtain sigmoid probabilities for predicates such as degraded, near_failure, nominal and a health embedding
- Pass probabilistic predicates and health scores to a MaintenanceSymbolicPlanner
- Generate feasible action sets by ranking actuators by risk reduction per unit cost and pruning via CP‑SAT style constraints
- Select optimal maintenance actions within budget and horizon
- Execute maintenance actions and update sensor data
- Iterate the loop
证据
Soft actuators lose 2-3% of its force output per thousand cycles due to elastomer fatigue
The symbolic planner receives predicate probabilities, not booleans
- 记录日期: 2026-10-10T16:40:51.036Z