Lifelong Multi-Agent Path Finding in A Dynamic Environment

Wan, Qian; Gu, Chonglin; Sun, Sankui; Chen, Mengxia; Huang, Hejiao; Jia, Xiaohua

doi:10.1109/icarcv.2018.8581181

Cited by 30 publications

(13 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In LMAPF, agents aim at continually planning paths online and maximizing the throughput (i.e., the average number of targets reached per timestep). To implement conventional baselines for LMAPF, we decompose the problem into a series of one-shot MAPF instances as is commonly done [23]. We select ODrM* (with inflation factors ε = 10), CBS, and iECBS as our bounded suboptimal planners [21], [20], [34] and use a timeout of 60s when using C++ implementations and 300s for python codes to remain consistent with other works in the field.…”

Section: B Lmapf Resultsmentioning

confidence: 99%

“…One of the common approaches to solve LMAPF involves stitching one-shot MAPF instances together by using a (usually complete, bounded-suboptimal) MAPF planner to recompute paths at each timestep at least one agent is assigned a new goal [4], [5], [23]. However, replanning time grows exponentially with the number of agents, and resources are wasted in the redundant computation of paths for agents whose goals are unaffected.…”

Section: B Lifelong Multi-agent Pathfindingmentioning

confidence: 99%

See 1 more Smart Citation

PRIMAL$_2$: Pathfinding Via Reinforcement and Imitation Multi-Agent Learning - Lifelong

Damani

Luo

Wenzel

et al. 2021

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

Multi-agent path finding (MAPF) is an indispensable component of large-scale robot deployments in numerous domains ranging from airport management to warehouse automation. In particular, this work addresses lifelong MAPF (LMAPF) -an online variant of the problem where agents are immediately assigned a new goal upon reaching their current one -in dense and highly structured environments, typical of real-world warehouse operations. Effectively solving LMAPF in such environments requires expensive coordination between agents as well as frequent replanning abilities, a daunting task for existing coupled and decoupled approaches alike. With the purpose of achieving considerable agent coordination without any compromise on reactivity and scalability, we introduce PRIMAL 2 , a distributed reinforcement learning framework for LMAPF where agents learn fully decentralized policies to reactively plan paths online in a partially observable world. We extend our previous work, which was effective in low-density sparsely occupied worlds, to highly structured and constrained worlds by identifying behaviors and conventions which improve implicit agent coordination, and enabling their learning through the construction of a novel local agent observation and various training aids. We present extensive results of PRIMAL 2 in both MAPF and LMAPF environments with up to 1024 agents and compare its performance to complete state-of-the-art planners. We experimentally observe that agents successfully learn to follow ideal conventions and can exhibit selfless coordinated maneuvers that maximize joint rewards. We find that not only does PRIMAL 2 significantly surpass our previous work, it is also able to perform on par and even outperform state-of-theart planners in terms of throughput.

show abstract

Section: B Lmapf Resultsmentioning

confidence: 99%

Section: B Lifelong Multi-agent Pathfindingmentioning

confidence: 99%

PRIMAL$_2$: Pathfinding Via Reinforcement and Imitation Multi-Agent Learning - Lifelong

Damani

Luo

Wenzel

et al. 2021

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

“…1(b), where each 3 × 3 cell has a hole in the middle, can be natively solved without performance degradation. Such settings find real-world applications in parcel sorting facilities in large warehouses [45,26]. For this parcel sorting setup, We fix the robot density at 2 9 and test ECBS, DDM, RTH and RTH-LBA on graphs with varying sizes.…”

Section: B Evaluation and Comparative Study Of Rthmentioning

confidence: 99%

“…In practice, solution optimality is also of key importance; yet optimally solving MRPP is generally NP-hard [42,50], even in planar [47] and grid settings [9]. MRPP algorithms find many important largescale applications, including, e.g., in warehouse automation for general order fulfillment [46], grocery order fulfillment [32], and parcel sorting [45]. Other application scenarios include formation reconfiguration [35], agriculture [6], object transportation [37], swarm robotics [36,21], to list a few.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Sub-1.5 Time-Optimal Multi-Robot Path Planning on Grids in Polynomial Time

Yu¹

2022

Preprint

View full text Add to dashboard Cite

It is well-known that graph-based multi-robot path planning (MRPP) is NP-hard to optimally solve. In this work, we propose the first low polynomial-time algorithm for MRPP achieving 1-1.5 asymptotic optimality guarantees on solution makespan (i.e., the time it takes to complete a reconfiguration of the robots) for random instances under very high robot density, with high probability. Specifically, on an m1 ×m2 gird, m1 ≥ m2, our RTH (Rubik Table with Highways) algorithm computes solutions for routing up to m 1 m 2 3 robots with uniformly randomly distributed start and goal configurations with a makespan of m1 + 2m2 + o(m1), with high probability. Because the minimum makespan for such instances is m1 + m2 − o(m1), also with high probability, RTH guarantees m 1 +2m 2 m 1 +m 2 optimality as m1 → ∞ for random instances with up to 1 3 robot density, with high probability. m 1 +2m 2 m 1 +m 2 ∈ (1, 1.5]. Alongside the above-mentioned key result, we also establish:(1) for completely filled grids, i.e., m1m2 robots, any MRPP instance may be solved in polynomial time under a makespan of 7m1 + 14m2, (2) for m 1 m 2 3 robots, RTH solves arbitrary MRPP instances with makespan of 3m1 + 4m2 + o(m1), (3) for m 1 m 2 2 robots, a variation of RTH solves a random MRPP instance with the same 1-1.5 optimality guarantee, and (4) the same m 1 +2m 2 m 1 +m 2 optimality guarantee holds for regularly distributed obstacles at 1 9 density together with 2m 1 m 2 9 randomly distributed robots; such settings directly map to real-world parcel sorting scenarios.Moreover, we have developed effective, principled heuristics that further improve the computed optimality of RTH algorithms. In extensive numerical evaluations, RTH and its variants demonstrate exceptional scalability as compared with methods including ECBS and DDM, scaling to over 450 × 300 grids with 45, 000 robots, and consistently achieves makespan around 1.5 optimal or better, as predicted by our theoretical analysis.• The same m1+2m2 m1+m2 optimality guarantee may be achieved on grids with up to m1m2 9 regularly distributed obstacles together with 2m1m2 9 robots using RTH (e.g., Fig. 1(b)).

show abstract