Inverse kinematics (IK) is a foundational primitive in robotics, mapping the task-space objectives expressed by higher-level planners to robot configurations that realize them. Traditional IK methods generally treat each query as an isolated pose-to-configuration problem, but planners often need a distribution over valid realizations that accounts for future reachability and preferences such as collision avoidance and stability. We introduce GenGenIK, a generative modeling approach that supports mobile bases, multiple end-effectors and targets, scene awareness, and preferences over robot behavior within a common interface. Experiments show that GenGenIK outperforms existing methods in both solve rate and runtime across tasks such as bimanual lift and flip-up, while allowing flexible control over the sampled distribution and remaining effective under partial scene observations. Website: https://icra-2027-gengenik.github.io/.
Figures & tables
Fig. 1: Example use cases of GenGenIK (under-table pickup, partially observable scenes, cluttered kitchens).
Fig. 2: Conditional architecture. Task poses and scene geometry form conditioning tokens, while the current candidate is represented by base and joint tokens. The decoder combines these with FK feedback and optional solution requests.
Robot
Method
Self-coll. (%) ↓
Pos. err. (mm) ↓
Rot. err. (deg) ↓
NFE ↓
Conv. (%) ↑
FR3 (1, 7)
IKD [ 7 ]
0.49
4.90 ± 0.284
1.05 ± 0.062
5
98.47
Ours
0.52
4.73 ± 0.275
0.78 ± 0.057
4.1
98.50
LEAP (4, 22) 6DoF-base
IKD [ 7 ]
2.85
1.73 ± 0.072
0.93 ± 0.033
5
80.95
Ours
1.18
1.14 ± 0.049
0.80 ± 0.034
2.4
95.81
G1 (4, 29)
IKD [ 7 ]
6.35
5.31 ± 0.156
1.48 ± 0.049
5
96.96
Ours
1.97
3.44 ± 0.159
0.80 ± 0.042
4.6
98.30
TABLE I: Scene-free point IK. Robots are specified as (number of end-effectors, degrees of freedom [DoF]). Position and rotation errors are mean ± standard error over 100 samples for each of 10,000 targets. NFE includes redraws. Reported baselines are from the respective papers (standard errors are estimated assuming 100 samples). Conv. measures per-sample convergence when seeding cuRobo with model samples, requiring position error below 0.1 mm and rotation error below 0.01 rad.
Method
Train
Test
S@1 (%)
S@8 (%)
WF@1 (%)
TTFV (ms) B1 B32
Ours
full
full
80.9
98.2
84.7
66.4
5.6
Ours (1-NFE)
full
full
71.2
96.8
75.6
12.1
2.4
Ours
full
none
16.7
52.3
17.6
156
11.8
Ours
full
wrong cloud
20.7
40.7
22.7
225
18.9
Ours (blind)
none
none
14.6
48.6
16.5
171
18.4
Ours (raw) + cuRobo
full
full (cuboids)
90.6
93.3
95.4
173
92.2
TABLE II: Scene-aware point IK on RB-Y1. Results use 1024 grasp problems from 252 held-out iTHOR and ProcTHOR scenes. Models are trained on full point clouds (full), partial point clouds (partial), or no scene data (blind). Ours uses a 32-NFE model with our refiner unless otherwise specified; cuRobo uses cuboids for full scenes and 2.5 cm voxels with unobserved space free for partial scenes. All samples are evaluated against the complete ground-truth scene. Columns specify success@1, success@8, and world-collision-free@1. TTFV uses rounds of K=8 samples.
Task
Method
Success@32 (%)
TTFV , B1 (ms)
Door/drawer (1, L= 3, 5)
cuRobo
89.3
170.5
Ours
100.0
22.7
Pushing (1, L= 2, 5)
cuRobo
92.0
167.3
Ours
98.7
20.2
Bimanual lift (2, L= 5)
cuRobo
42.7
940.0
Ours
92.7
36.1
TABLE III: Path IK on RB-Y1. Tasks are specified as (number of end-effectors, path length(s)). Ours uses 1-NFE reflow students and our refiner. Each task family uses 150 problems from 150 distinct scenes with 32 draws per problem. TTFV uses rounds of K=32 samples.
q1 source
Push
Door
Drawer
Lift
Flip
Under-table
cuRobo, H1 only
26.0
18.5
47.4
41.2
5.4
7.3
Ours, full path
76.2
56.2
87.0
52.9
29.1
47.9
TABLE IV: Fraction (%) of valid first-target configurations that admit a valid cuRobo continuation, evaluated on 100 tasks per task family.
Fig. 3: Preference-conditioned IK in the RB-Y1 bimanual lift task. Left: requesting a base stand-off ≤0.45 m shifts generated base placements inward relative to unconditional samples, where stand-off is measured from the anchor end-effector target (black box) to the base axle. Right: requesting right-elbow height ≥+0.15 m and left-elbow height ≤−0.10 m yields an asymmetric posture, where elbow height is the world-frame vertical offset from the shoulder.
Parameter
Models
Value
Shared
Soft-bound margin μ
All
2∘
S1 weight
All
0.05
Minimum time ϵ
Except under-table
0.01
Under-table
0.02
Teacher training
TABLE VI: Selected training hyperparameters. Teacher auxiliary-loss weights are relative to the flow-matching loss.
Inverse kinematics maps desired robot or articulated-body motions to joint configurations while handling constraints, redundancy, and multiple solutions.
Neural models can learn to generate various solutions to the inverse kinematics problem from data, but are usually limited to a single robot. We present MorphIK, a flow-matching model that solves inverse kinematics for revolute-joint-based kinematic chains it has never seen during training. The model uses a transformer architecture to encode the robot's morphology along with the target pose. This encoding then conditions a flow-matching head that generates poses from noise. Trained on purely synthetic data from procedurally generated robots, the model reaches a precision of about 5 cm on unseen real-world robots with 6 to 9 Degrees of Freedom. For higher precision, the model serves as an excellent Prior for further optimization algorithms, reducing error to less than 1 cm after a single step of Damped Least Squares optimization and to sub-1 mm error after 3 steps in most cases. Building on flow matching's generative capabilities to produce highly diverse outputs, our model can efficiently sample the robot's null space, providing a wide variety of configurations for the same pose. Thus, overall, MorphIK allows learning and generalizing neural inverse kinematics for a multitude of known and unknown robots.
Lennart Clasmeier, Jan Gerrit Habekost, Cornelius Weber +1
Knowledge Technology Group, Department of Informatics, University of Hamburg, Hamburg, Germany
Inverse kinematics (IK) remains a critical bottleneck for real-time robot manipulation. Classical numerical solvers achieve high geometric precision but often suffer from discontinuous branch switching and unstable behavior near kinematic singularities during closed-loop deployment. Meanwhile, learned IK approaches frequently struggle to balance spatial accuracy, motion smoothness, and real-time efficiency, particularly when trained on noisy human teleoperation data. We present MimicIK, a real-time generative inverse kinematics framework that learns smooth and robust joint-space motion priors from teleoperation demonstrations through conditional flow matching. Given the current joint configuration and a target end-effector pose, MimicIK predicts continuous delta-joint commands using an efficient two-step iterative refinement process based on a Minimal Iterative Policy (MIP) backbone. To enforce physical consistency, we further introduce an FK consistency loss, a differentiable forward-kinematics regularization that penalizes task-space deviations from the target pose during training. We evaluate MimicIK on a real-world 6-DOF robot dataset containing 8,848 teleoperation demonstrations. MimicIK achieves a mean position error of 4.65 mm, a 10 mm success rate of 92.01%, and a trajectory spike rate of only 7.99%. Compared with a UNet diffusion baseline, our method improves both spatial accuracy and motion smoothness while reducing inference latency from 21.66 ms to 6.74 ms. Furthermore, unlike deterministic MLP baselines that catastrophically diverge under out-of-distribution deployment, MimicIK remains stable near singular configurations and enables robust 20 Hz real-time control on deployment hardware.
Inverse kinematics (IK) is a fundamental problem in robotics, requiring the generation of joint configurations that satisfy target end-effector poses. Existing approaches often struggle to generalize across diverse robot morphologies and to effectively model the multi-modal nature of IK, particularly in articulated systems with multiple kinematic branches. In this work, we propose GraphDiff-IK, a structure-aware graph diffusion framework for inverse kinematics. Specifically, we represent the robot as a kinematic graph constructed from the robot URDF, where nodes correspond to actuated joints and edges encode kinematic dependencies. Building upon this representation, we formulate IK as a conditional graph diffusion process that directly generates joint configurations on the robot graph. To better capture structural dependencies in articulated systems, we further introduce a structure-aware graph reasoning framework with hierarchical stage-wise message passing and torso-aware conditioning for multi-branch robots. In addition, we incorporate noisy forward kinematics feedback and task-space supervision to improve geometric consistency during denoising. The proposed framework provides a unified formulation that naturally supports single-arm robots, dual-arm systems, and articulated robots with torso or waist structures. Extensive experiments on diverse robotic platforms demonstrate that the proposed method achieves accurate and stable IK performance while preserving the ability to generate multiple feasible solutions for redundant robotic systems.