min-err-trajs-vnncomp-benchmarks (rl_benchmarks)

Benchmarks for vnncomp 2022, generated from work on Minimum-Error Trajectories

Reinforcement Learning

This set of benchmarks is related to classical reinforcement learning problems.

Cartpole

This benchmark is the standard inverted pendulum that must be stabilized

Dubins Rejoin

This benchmark is the problem of making an unmanned aerial vehicle to follow a manned one acting as the lead

Lunar Lander

This benchmark is the rocket trajectory optimization for landing correctly in a given place a two-motor aerial vehicle

--- List of all rl_benchmarks [fullycon_net] networks (From : vnncomp2022_benchmarks )---

cartpole.onnx

Number of parameters: 4610 
Node types: ['Gemm' 'Flatten' 'Relu']

dubinsrejoin.onnx

Number of parameters: 70152 
Node types: ['Relu' 'Add' 'MatMul']

lunarlander.onnx

Number of parameters: 4996 
Node types: ['Gemm' 'Flatten' 'Relu']