卓汉逵- 如何让大语言模型协助经典智能规划.pdf

1、如何让大模型辅助智能规划卓汉逵https:/www.xplan-lab.org动机问题：质量不稳定，效率低目标：完成工序，提高质量、稼动率特点：动态，混料，柔性，实时动机Long term benefitLarge scale orders Large scale resource schedulingLots of manufacturing stepsComplicated supply chainLabor intensive 1.知识获取=规划领域模型：质量和规模 2.规划求解=动作序列：效率和质量智能规划问题 Symbolic planning:domain models synth

2、esizing plans to transit initial states to goals Different from classical control and classification problems,the solutions are complex and must be optimized in multi-step reasoning.location 1location 2location 1location 2s1s3s4takeputlocation 1location 2location 1location 2s0s2s5move1puttakemove1mo

3、ve1move2loadunloadmove2move2location 1location 2location 1location 2初始状态目标状态 智能规划问题：(O,s0,g)O是动作模型的集合s0是初始状态g是目标状态 智能规划问题的解 规划：=a1,a2,an 规划是智能规划问题的解，如果该规划可被执行，并最终实现目标g：(s0,a1)=s1 (s1,a2)=s2 (sn1,an)=snsn satisfies g智能规划问题 三个不同的解：take(crane1,loc1,c3,c1,p1),move(r1,loc2,loc1),move(r1,loc1,loc2),move(r

4、1,loc2,loc1),load(crane1,loc1,c3,r1),move(r1,loc1,loc2)take(crane1,loc1,c3,c1,p1),move(r1,loc2,loc1),load(crane1,loc1,c3,r1),move(r1,loc1,loc2)move(r1,loc2,loc1),take(crane1,loc1,c3,c1,p1),load(crane1,loc1,c3,r1),move(r1,loc1,loc2)6/43s1=attached(p1,loc1),in(c1,p1),in(c3,p1),top(c3,p1),on(c3,c1),on(

5、c1,pallet),attached(p2,loc1),in(c2,p2),top(c2,p2),on(c2,pallet),belong(crane1,loc1),empty(crane1),adjacent(loc1,loc2),adjacent(loc2,loc1),at(r1,loc2),occupied(loc2),unloaded(r1)r1crane1p2p1c2c1c3loc1loc2 States/actions are represented by propositions form State space explosively increase w.r.t.objec

6、ts Action space explosively increase w.r.t.objectsLarge scale problem?智能规划问题take c3move r1take c2ForwardSearch(O,s0,g)s=s0=nullloop if s 满足 g，则返回 E=a|a 是 O 的实例，precond(a)在 s 中为真 if E=，则返回失败 选择从 E 中选择动作 a s=(,)=规划求解：前向/后向搜索g0g1g2g3a1a2a3g4g5s0a4a5BackwardSearch(O,s0,g)=nullloop if s0 满足 g，则返回 A=a|a 是