From 0b3cd3ab2fdb5cf22a2c5799875c23668a3266fe Mon Sep 17 00:00:00 2001 From: wertyuilife Date: Mon, 30 Mar 2026 19:18:31 +0800 Subject: [PATCH] align hip_pos_penalty reward with isaacgym (switch to hip_pos_penalty_l1). --- source/robot_lab/robot_lab/tasks/go2/env_cfg.py | 4 ++-- source/robot_lab/robot_lab/tasks/go2/mdp/rewards.py | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/source/robot_lab/robot_lab/tasks/go2/env_cfg.py b/source/robot_lab/robot_lab/tasks/go2/env_cfg.py index a430638..5dbc7a1 100644 --- a/source/robot_lab/robot_lab/tasks/go2/env_cfg.py +++ b/source/robot_lab/robot_lab/tasks/go2/env_cfg.py @@ -431,8 +431,8 @@ class RewardsCfg: weight=-1.0, params={"sensor_cfg": SceneEntityCfg("contact_forces", body_names=".*_thigh|.*_calf"), "threshold": 5.0}, ) - hip_pos_penalty = RewTerm( - func=mdp.hip_pos_penalty, + hip_pos_penalty_l1 = RewTerm( + func=mdp.hip_pos_penalty_l1, weight=-0.05, params={ "command_name": "base_velocity", diff --git a/source/robot_lab/robot_lab/tasks/go2/mdp/rewards.py b/source/robot_lab/robot_lab/tasks/go2/mdp/rewards.py index d23ad9b..ae974ae 100644 --- a/source/robot_lab/robot_lab/tasks/go2/mdp/rewards.py +++ b/source/robot_lab/robot_lab/tasks/go2/mdp/rewards.py @@ -534,7 +534,7 @@ def flat_orientation_l2(env: ManagerBasedRLEnv, asset_cfg: SceneEntityCfg = Scen return reward -def hip_pos_penalty( +def hip_pos_penalty_l1( env: ManagerBasedRLEnv, command_name: str, asset_cfg: SceneEntityCfg, @@ -547,7 +547,7 @@ def hip_pos_penalty( command = env.command_manager.get_command(command_name)[:, [1, 2]] cmd_large = torch.any(torch.abs(command) > command_threshold, dim=1) running_reward = torch.linalg.norm( - (asset.data.joint_pos[:, asset_cfg.joint_ids] - asset.data.default_joint_pos[:, asset_cfg.joint_ids]), dim=1 + (asset.data.joint_pos[:, asset_cfg.joint_ids] - asset.data.default_joint_pos[:, asset_cfg.joint_ids]), dim=1, ord=1 ) reward = torch.where( cmd_large,