Constraints

We also provide a util class to compute the constraints. The ConstraintList is a collection of the Constraint instances that is available in env_info. Here is an simple example explaining how to use the constraint functions.

Here is an example of how to access the constraints:

from air_hockey_challenge.framework import AirHockeyChallengeWrapper

env = AirHockeyChallengeWrapper('3dof-hit')

# Setup the environment
obs = env.reset()

env_info = env.env_info

# Get the keys of the available constraint
print(env_info['constraints'].keys())
# dict_keys(['joint_pos_constr', 'joint_vel_constr', 'ee_constr'])

# Get the joint position and velocity from the observation
q = obs[env_info['joint_pos_ids']]
dq = obs[env_info['joint_vel_ids']]

# Get a dictionary of the constraint functions {"constraint_name": ndarray}
c = env_info['constraints'].fun(q, dq)

# Get a dictionary of the constraint jacobian {"constraint_name": 2d ndarray}
jac = env_info['constraints'].jacobian(q, dq)

# Get value of the constraint function by name
c_ee = env_info['constraints'].get('ee_constr').fun(q, dq)

# Get jacobian of the constraint function by name
jac_vel = env_info['constraints'].get('joint_vel_constr').jacobian(q, dq)

We also list all of the available constraint here:

Class Name	Key	Output Dim	Description
JointPositionConstraint	“joint_pos_constr”	2 * num_joints	\(q_l < q_{cmd} < q_u\)
JointVelocityConstraint	“joint_vel_constr”	2 * num_joints	\(\dot{q}_l < \dot{q}_{cmd} < \dot{q}_u\)
EndEffectorConstraint	“ee_constr”	5	\(l_x < x_{ee},\) \(l_y < y_{ee} < u_y,\) \(z_{ee} > \mathrm{table\,height - tolerance}\), \(z_{ee} < \mathrm{table\, height + tolerance}\).
LinkConstraint (7DoF Robot Only)	“link_constr”	2	\(z_{elbow} > 0.25\), \(z_{wrist} > 0.25\)