The Holy Grail of Multi-Robot Planning: Learning to Generate Online-Scalable Solutions from Offline-Optimal Experts

@inproceedings{Prorok2022TheHG,
  title={The Holy Grail of Multi-Robot Planning: Learning to Generate Online-Scalable Solutions from Offline-Optimal Experts},
  author={Amanda Prorok and Jan Blumenkamp and Qingbiao Li and Ryan Kortvelesy and Zhe Liu and Ethan Stump},
  booktitle={AAMAS},
  year={2022}
}
Many multi-robot planning problems are burdened by the curse of dimensionality, which compounds the difficulty of applying solutions to large-scale problem instances. The use of learning-based methods in multi-robot planning holds great promise as it enables us to offload the online computational burden of expensive centralized, yet optimal solvers, to an offline learning procedure. The hope is that by training a policy to copy an optimal pattern generated by a small-scale (centralized) system… 

Figures from this paper

A Framework for Real-World Multi-Robot Systems Running Decentralized GNN-Based Policies

TLDR
This work creates a framework based on ROS2 and elaborate its details in this paper, and presents first-of-a-kind results that show successful real-world deployment of GNN-based policies on a decentralized multi-robot system relying on Adhoc communication.

Synthesizing Decentralized Controllers With Graph Neural Networks and Imitation Learning

TLDR
GNNs learn appropriate decentralized controllers by means of imitation learning, leverage their permutation invariance properties to successfully scale to larger teams and transfer to unseen scenarios at deployment time, and the problems of flocking and multi-agent path planning are explored.

A Critical Review of Communications in Multi-Robot Systems

TLDR
A critical view of the way robotic algorithms and their networking systems have evolved is presented, and the case for a more synergistic approach to co-design and co-optimization of robots and the networks they employ is made.

References

SHOWING 1-10 OF 78 REFERENCES

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

TLDR
PRIMAL is introduced, a distributed reinforcement learning framework for LMAPF where agents learn fully decentralized policies to reactively plan paths online in a partially observable world and performs comparably to state-of-the-art planners in terms of makespan and throughput.

Graph Neural Networks for Decentralized Multi-Robot Path Planning

TLDR
A combined model is proposed that automatically synthesizes local communication and decision-making policies for robots navigating in constrained workspaces that shows its capability to generalize to previously unseen cases.

Mobile Robot Path Planning in Dynamic Environments Through Globally Guided Reinforcement Learning

TLDR
This work introduces a globally guided reinforcement learning approach (G2RL), which incorporates a novel reward structure that generalizes to arbitrary environments and applies G2RL to solve the multi-robot path planning problem in a fully distributed reactive manner.

A Framework for Real-World Multi-Robot Systems Running Decentralized GNN-Based Policies

TLDR
This work creates a framework based on ROS2 and elaborate its details in this paper, and presents first-of-a-kind results that show successful real-world deployment of GNN-based policies on a decentralized multi-robot system relying on Adhoc communication.

Graph Policy Gradients for Large Scale Robot Control

TLDR
This paper proposes a new algorithm called Graph Policy Gradients (GPG) that exploits the underlying graph symmetry among the robots that is able to scale better than existing reinforcement methods that employ fully connected networks.

VR-Goggles for Robots: Real-to-Sim Domain Adaptation for Visual Control

TLDR
This letter proposes a simple yet effective shift loss that is agnostic to the downstream task, to constrain the consistency between subsequent frames which is important for consistent policy outputs, and validate the shift loss for artistic style transfer for videos and domain adaptation.

Learned Sampling Distributions for Efficient Planning in Hybrid Geometric and Object-Level Representations

TLDR
This work shows that learning a sampling distribution that incorporates both geometric information and explicit, object-level semantics for sampling-based planners enables efficient planning at longer horizons in partially-known environments.

Reinforcement Learning for Solving the Vehicle Routing Problem

TLDR
This work presents an end-to-end framework for solving the Vehicle Routing Problem (VRP) using reinforcement learning, and demonstrates how this approach can handle problems with split delivery and explore the effect of such deliveries on the solution quality.

Message-Aware Graph Attention Networks for Large-Scale Multi-Robot Path Planning

TLDR
The Message-Aware Graph Attention neTwork (MAGAT) is based on a key-query-like mechanism that determines the relative importance of features in the messages received from various neighboring robots and is able to achieve a performance close to that of a coupled centralized expert algorithm.

Decentralized path planning for multi-agent teams with complex constraints

TLDR
A version of the algorithm is introduced, called Cooperative DMA-RRT, which introduces a cooperation strategy that allows an agent to modify its teammates’ plans in order to select paths that reduce their combined cost, which improves team performance and avoids certain common deadlock scenarios.
...