curobo.opt package

Optimization module containing several numerical solvers.

Base for an opimization solver is at opt_base.Optimizer. cuRobo provides two base classes for implementing two popular ways to optimize, (1) using particles with particle.particle_opt_base.ParticleOptBase and (2) using Newton/Quasi-Newton solvers with newton.newton_base.NewtonOptBase. newton.newton_base.NewtonOptBase contains implementations of several line search schemes. Note that these line search schemes are approximate as cuRobo tries different line search magnitudes in parallel and chooses the largest that satisfies line search conditions.

Subpackages

• curobo.opt.newton package
  • Submodules
    • curobo.opt.newton.lbfgs module
      • jit_lbfgs_compute_step_direction
      • jit_lbfgs_update_buffers
      • LBFGSOptConfig
      • LBFGSOpt
    • curobo.opt.newton.newton_base module
      • LineSearchType
      • NewtonOptConfig
      • NewtonOptBase
      • get_x_set_jit
      • _armijo_line_search_tail_jit
      • _wolfe_search_tail_jit
      • scale_action
      • check_convergence
• curobo.opt.particle package
  • Submodules
    • curobo.opt.particle.parallel_es module
      • ParallelESConfig
      • ParallelES
      • calc_exp
      • compute_es_mean
    • curobo.opt.particle.parallel_mppi module
      • BaseActionType
      • CovType
      • ParallelMPPIConfig
      • ParallelMPPI
      • jit_calculate_exp_util
      • jit_calculate_exp_util_from_costs
      • jit_compute_total_cost
      • jit_diag_a_cov_update
      • jit_blend_cov
      • jit_blend_mean
      • jit_mean_cov_diag_a
    • curobo.opt.particle.particle_opt_base module
      • SampleMode
      • ParticleOptInfo
      • ParticleOptConfig
      • ParticleOptBase
    • curobo.opt.particle.particle_opt_utils module
      • SquashType
      • scale_ctrl
      • get_stomp_cov
      • get_stomp_cov_jit
      • gaussian_logprob
      • gaussian_logprobgrad
      • gaussian_entropy
      • gaussian_kl
      • cost_to_go
      • cost_to_go_np
      • matrix_cholesky
      • batch_cholesky

Submodules

• curobo.opt.opt_base module
  • OptimizerConfig
    • OptimizerConfig.d_action
    • OptimizerConfig.action_lows
    • OptimizerConfig.action_highs
    • OptimizerConfig.action_horizon
    • OptimizerConfig.horizon
    • OptimizerConfig.n_iters
    • OptimizerConfig.cold_start_n_iters
    • OptimizerConfig.rollout_fn
    • OptimizerConfig.tensor_args
    • OptimizerConfig.use_cuda_graph
    • OptimizerConfig.store_debug
    • OptimizerConfig.debug_info
    • OptimizerConfig.n_problems
    • OptimizerConfig.num_particles
    • OptimizerConfig.sync_cuda_time
    • OptimizerConfig.use_coo_sparse
    • OptimizerConfig.create_data_dict
  • Optimizer
    • Optimizer.optimize
    • Optimizer.update_params
    • Optimizer.reset
    • Optimizer.update_nproblems
    • Optimizer.get_nproblem_tensor
    • Optimizer.reset_seed
    • Optimizer.reset_cuda_graph
    • Optimizer.reset_shape
    • Optimizer.get_all_rollout_instances
    • Optimizer._optimize
    • Optimizer._shift
    • Optimizer._update_problem_kernel
    • Optimizer.create_data_dict
    • Optimizer.d_action
    • Optimizer.action_lows
    • Optimizer.action_highs
    • Optimizer.action_horizon
    • Optimizer.horizon
    • Optimizer.n_iters
    • Optimizer.cold_start_n_iters
    • Optimizer.rollout_fn
    • Optimizer.tensor_args
    • Optimizer.use_cuda_graph
    • Optimizer.store_debug
    • Optimizer.debug_info
    • Optimizer.n_problems
    • Optimizer.num_particles
    • Optimizer.sync_cuda_time
    • Optimizer.use_coo_sparse