Compared　to　human-driven　vehicles　(HDVs),　connected　and　autonomous　vehicles　(CAVs)　can　drive　closer　to　each　other　to　enhance　link　capacity.　Thereby,　they　have　great　potential　to　mitigate　traffic　congestion.　However,　the　presence　of　HDVs　in　mixed　traffic　can　significantly　reduce　the　effects　of　CAVs　on　link　capacity,　especially　when　the　proportion　of　HDVs　is　high.　To　address　this　problem,　this　study　seeks　to　control　the　HDV　flow　using　the　autonomous　vehicle/toll　(AVT)　lanes　introduced　by　Liu　and　Song　(2019).　The　AVT　lanes　grant　free　access　to　CAVs　while　allowing　HDVs　to　access　by　paying　a　toll.　To　find　the　optimal　toll　rates　for　the　AVT　lanes　to　improve　the　network　performance,　first,　this　study　proposes　a　multiclass　traffic　assignment　problem　with　elastic　demand　(MTA-ED　problem)　to　estimate　the　impacts　of　link　tolls　on　equilibrium　flows.　It　not　only　enhances　behavioral　realism　for　modeling　the　route　choices　of　HDV　and　CAV　travelers　by　considering　their　knowledge　level　of　traffic　conditions　but　also　captures　the　elasticity　of　both　HDV　and　CAV　demand　in　response　to　the　changes　in　the　level　of　service　induced　by　the　tolls　on　AVT　lanes.　Thereby,　it　better　estimate　the　equilibrium　network　flows　after　the　tolls　are　deployed.　Then,　two　categories　of　optimal　toll　design　problems　are　formulated　according　to　whether　the　solution　of　the　HDV　route　flows,　CAV　link　flows　and　corresponding　origin?destination　demand　of　the　proposed　MTA-ED　problem　is　unique　or　not.　To　solve　these　optimal　toll　design　problems,　this　study　proposes　a　revised　method　of　feasible　direction.　It　linearizes　the　anonymous　terms　in　the　upper-level　problem　by　leveraging　the　analytical　sensitivity　analysis　results　of　the　lower-level　MTA-ED　problem.　This　algorithm　is　globally　convergent　on　the　condition　that　the　MTA-ED　problem　has　a　unique　solution.　It　can　also　be　leveraged　to　solve　optimal　toll　design　problems　when　the　MTA-ED　problem　has　multiple　solutions.　Numerical　application　found　that　due　to　disruptive　effects　on　link　capacity,　using　HDVs　may　significantly　reduce　the　network　performance　such　as　customer　surplus　and　total　travel　demand.　The　proposed　method　can　assist　different　stakeholders　to　find　the　optimal　toll　rates　for　HDVs　on　AVT　lanes　to　maximize　the　network　performance　under　mixed　traffic　environments.