For　the　collaborative　formation　control　problem　of　a　multi-snake　robot　with　uncertain　sideslip　and　lumped　unknowns,　this　study　proposed　collaborative　line-of-sight　(LOS)　guidance　and　consensus-based　path　variable,　which　enables　snake　robots　to　track　multiple　parametric　paths　in　a　fixed　formation.　Meanwhile,　the　improved　finite-time　sideslip　observer　effectively　compensates　for　the　kinematic　deviation,　thereby　improving　the　accuracy　of　the　multi-snake　robot＇s　path　tracking.　A　robust　adaptive　controller　based　on　an　integral　sliding　mode　manifold　is　developed　to　eliminate　external　disturbances.　The　model　uncertainty　and　time-varying　disturbances　are　comprehensively　involved　and　incorporated　as　lumped　unknowns　in　dynamics.　The　prediction-based　RBF　neural　network　estimators　are　derived　to　approximate　unknowns　and　compensate　controllers.　The　Lyapunov　theory　is　adopted　to　prove　the　stability　of　formation　errors　for　snake　robots　in　the　closed-loop　system.　Simulation　and　experiment　results　show　the　practicability　and　superiority　of　this　work.　Note　to　Practitioners-This　paper　is　mainly　motivated　to　study　a　cooperative　formation　control　framework　suitable　for　snake　robots＇　dynamics,　especially　for　the　application　of　resisting　model　uncertainty　and　external　interference.　In　practice,　the　collaborative　line　of　sight　guidance　law　is　employed　to　generate　and　adjust　the　reference　direction　and　speed　of　robots　in　real　time,　and　then　the　method　is　applied　to　the　joint　torque　controller　to　guide　robots　to　track　the　desired　parameterized　path.　Meanwhile,　the　improved　finite　time　sideslip　observer　corrects　the　deviation　of　the　guidance　law,　ensuring　the　observation　accuracy　while　reducing　the　computational　time.　During　the　control　process,　for　highly　coupled　and　complex　snake　robot　models,　solving　external　disturbances　and　model　uncertainties　remains　a　challenge.　A　neural　network　based　state　estimator　is　used　to　handle　these　unknowns　to　pre-compensate　for　underrotation/overrotation　of　the　joint　steering　gear　when　disturbed.　Simulation　and　experimental　results　validate　that　the　proposed　scheme　can　enable　multiple　snake　robots　to　synergistically　perform　tasks,　expressing　high　potential　in　unknown　environments.　In　the　future,　we　plan　to　design　a　three-dimensional　spiral　gait　instead　of　the　two-dimensional　one　used　in　this　study.　With　this　change,　we　will　further　improve　the　usability　of　snake　robots　in　real-world　scenarios.