Organizations: Northeastern University, Boston, MA, USA · Syracuse University, Syracuse, NY, USA · University of California, Irvine, CA, USA · The George Washington University, Washington, DC, USA
Safety filters for heterogeneous robot teams fail because of missing constraints, overestimated braking capabilities and conflicts between individually satisfiable constraints. We present OntoCBF, which compiles a domain ontology with a class taxonomy, typed relations, safety axioms and a dynamics profile per robot class into a control barrier function filter with one quadratic program per agent. Description-logic entailment supplies the constraints of any team online, including those that follow from chains of relations and from classes defined by what a robot carries, so the existing safety axioms cover new robot classes, payload classes and combinations. We prove that a subclass may reuse the braking capability of its parent whenever it can produce every velocity of the parent, a property separate from subsumption. For box-bounded double integrators this property is also necessary, and every loss of safety then begins with a conflict. Once three robots interact, the safe set of pairwise barriers contains states from which a collision is unavoidable, and such states persist in larger teams under any fixed capability reduction. In a simulated warehouse where humans ignore the robots, a filter written for the design classes violates safety in 89-100 % of the episodes with unseen classes. In all scenarios, including one with 32 robots, OntoCBF violates safety in 20-70 % of the episodes, each time after a conflict, with a median deepest intrusion of 10-17 % of the keep-out distance. A multi-agent PPO policy trained with the filter in the loop lowers the rate of infeasible quadratic programs, nearly all of them conflicts, by 18-41 %.
Figures & tables
Figure 1: (a) OntoCBF. (b) Robot classes with per-axis acceleration limits aˉ (m/s 2 ), and payload classes. LightDrone (green) is a dynamic subtype of Drone . HeavyDrone (red) brakes less strongly and Forklift has no design ancestor, so both receive their own capability. These classes and Chemical appear only in tests.
Line of work
Team
Het.
Sem.
Spec.
Switch
New
Reuse
Learn
Multi-robot CBF filters ( Wang et al., 2017 ; Kim et al., 2026b ; Sah and Keshavan, 2026 )
∘
✓
✗
✗
∘
✗
✗
✗
Learned graph CBFs ( Zhang et al., 2025b ; Yang et al., 2026 ; Kim et al., 2026a )
∘
✓
✗
✗
∘
∘
✗
✓
Safe MARL and shields ( Gu et al., 2023 ; ElSayed-Aly et al., 2021 ; Chatterji and Acar, 2025 )
∘
✗
∘
∘
✗
✗
✗
✓
Logic compiled into CBFs ( Lindemann and Dimarogonas, 2019 ; Molnar and Ames, 2023 )
∘
✗
∘
✓
✓
✗
✗
✗
Constraints from (V)LMs ( Brunke et al., 2025 ; Ravichandran et al., 2026 )
✗
✗
✓
✗
∘
∘
✗
∘
Ontologies for robot safety ( Wu et al., 2026 ; van der Ploeg et al., 2023 ; Fatima and Tolg, 2026 )
✗
✗
✓
✓
✗
∘
✗
∘
Table 1: Related work compared on team certificate, class-specific dynamics, typed relational constraints, formal specification, guarantee across switches, unseen types without redesign, certificate-reuse condition and learning in the loop (✓ addressed, ∘ partly, ✗ absent).
Episodes with a violation (%)
Single-constraint failures
Scenario
Flat
Flat-cons
No-chain
Naive
Untyped
No-act.
OntoCBF
MAPPO
Flat
Naive
OntoCBF
W0 (design)
20
20
100
22
16
74
22
12
0
0
0
W1 (unseen robots)
89
17
100
23
17
72
20
11
182
266
0
W2 (unseen payload)
99
20
100
22
16
74
22
12
0
0
0
W3 (unseen combination)
100
28
100
40
37
96
40
26
0
0
0
W4 ( N=32 )
67
67
100
70
69
98
70
65
0
0
0
Table 2: Episodes with a violation (left, 100 % for None everywhere) and single-constraint failures summed over episodes (right, zero for all other methods), 100 seeds per cell, all with the PD controller except MAPPO, which runs behind OntoCBF (3 training runs pooled).
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
Class
Parent
Role in the experiments
aˉ
vˉ
Drone
Aerial
design class
2.0
1.5
LightDrone
Drone
test only, ⪯dynDrone , inherits
3.0
1.5
HeavyDrone
Drone
test only, ⪯dynDrone , own capability
1.0
1.5
Rover
Ground
design class
1.2
1.0
Carrier
Ground
design class
1.0
0.8
Forklift
Ground
test only, no design ancestor
1.2
1.0
Appendix
Table 3: Class profiles with the per-axis acceleration limit aˉ in m/s 2 and the nominal speed vˉ of the task controller in m/s. Robot–human constraints need βC>aˉh , which holds for every robot class.
Figure 2: Scenario W3 (a LightDrone carries a Chemical and a Drone a HazardousPayload ), same seed under OntoCBF ( t=10 s) and under Flat at its first hazard violation ( t=2.6 s). For a drone that carries a hazardous payload, Flat falls back to its default payload keep-out and keeps the payload only 0.8 m from people.
Figure 3: Stress test of the capability assignment for HeavyDrone (Section 6.2 ) in head-on encounters with a human, all starting inside the safe set of the true capability. Under Naive the HeavyDrone inherits the capability of Drone , and OntoCBF assigns it its own capability. The left panel shows the encounters with a single-constraint failure and the right panel those with a violation.
Episodes with a violation (%)
Naive − OntoCBF, per episode
Single-constraint
Deliveries / courier
γ
Naive
OntoCBF
McNemar p
violating steps
aerial–human
failures (Naive)
Naive
OntoCBF
1
23
20
0.63
+0.8 [ −2.6 , +4.3 ]
−0.2 [ −1.5 , +1.0 ]
266
1.79
1.64
3
27
21
0.26
+1.9 [ −0.6 , +4.3 ]
+3.1 [ +0.4 , +6.1 ]
907
2.00
1.87
10
44
21
< 0.001
+2.9 [ −0.6 , +6.3 ]
+2.3 [ −3.1 , +7.3 ]
2223
2.13
2.02
Appendix
Table 4: Naive and OntoCBF in W1 (unseen robots) as the class- K gain grows, with the PD controller and 100 paired seeds ( γ=1 is the main setting). Differences carry paired 95 % bootstrap intervals, and p comes from McNemar’s exact test on episodes with a violation. OntoCBF has zero single-constraint failures at every gain.
Scenario
None
Flat
Flat-cons
No-chain
Naive
Untyped
No-act.
OntoCBF
W0 (design)
100 [96, 100]
20 [13, 29]
20 [13, 29]
100 [96, 100]
22 [15, 31]
16 [10, 24]
74 [65, 82]
22 [15, 31]
W1 (unseen robots)
100 [96, 100]
89 [81, 94]
17 [11, 26]
100 [96, 100]
23 [16, 32]
17 [11, 26]
72 [63, 80]
20 [13, 29]
W2 (unseen payload)
100 [96, 100]
99 [95, 100]
20 [13, 29]
100 [96, 100]
22 [15, 31]
16 [10, 24]
74 [65, 82]
22 [15, 31]
W3 (unseen combination)
100 [96, 100]
100 [96, 100]
28 [20, 37]
100 [96, 100]
40 [31, 50]
37 [28, 47]
96 [90, 98]
40 [31, 50]
W4 ( N=32 )
100 [96, 100]
67 [57, 75]
67 [57, 75]
100 [96, 100]
70 [60, 78]
69 [59, 77]
98 [93, 99]
70 [60, 78]
Appendix
Table 5: Episodes with a violation (%) in the warehouse with 95 % Wilson intervals, PD controller, 100 seeds per cell.
Scenario
Without
Measure
Ablation
OntoCBF
W0 (design)
property chain
payload+hazard viol.
500.0
5.2
W1 (unseen robots)
robot taxonomy
aerial–human viol.
150.7
1.9
W1 (unseen robots)
dynamic-subtype test
single-constraint failures (total)
266
0
W2 (unseen payload)
payload taxonomy
hazard–human viol.
360.1
4.4
W3 (unseen combination)
defined class
hazard–human viol.
835.0
28.2
W3 (unseen combination)
activation check
pickups into a violation (total)
217
0
Appendix
Table 7: Ablations that remove one element of OntoCBF, each in the scenario that needs the element. Violations are counted in constraint-steps per episode, and single-constraint failures and pickups into a violation are summed over 100 episodes.
Scenario
No robot tax.
No payload tax.
No defined class
OntoCBF
W0 (design)
=
=
=
22
W1 (unseen robots)
100
=
=
20
W2 (unseen payload)
=
99
=
22
W3 (unseen combination)
79
100
100
40
W4 ( N=32 )
=
=
=
70
Appendix
Table 8: Episodes with a violation (%) under single-element ablations where the element is needed, PD controller, 100 seeds. The sign = marks scenarios in which the ablation acts exactly as OntoCBF, seed by seed, and all three ablations have zero single-constraint failures.
Figure 4: Episodes with a violation against deliveries per courier, 100 seeds per point and 3 pooled training runs for MAPPO. Points further down and to the right are better.
Scenario
Deliveries/courier
QP infeas. (%)
Filter (ms/step)
Events/ep.
Reasoning (ms/event)
None
OntoCBF
W0 (design)
2.67
1.76
1.2
1.40
268
0.91
W1 (unseen robots)
2.83
1.64
1.1
1.42
267
0.91
W2 (unseen payload)
2.67
1.76
1.2
1.40
268
0.91
W3 (unseen combination)
2.67
1.43
1.6
1.58
262
0.93
W4 ( N=32 )
2.67
1.80
2.2
6.10
774
4.60
Appendix
Table 9: Deliveries per courier in the warehouse without a filter and with OntoCBF, and computational cost of OntoCBF with the PD controller on one CPU core.
Onsets after a conflict
Base rate (agent-steps)
Scenario
Onsets
within 1 s
within 2 s
all
close encounter
W0 (design)
45
45
45
2.0 %
3.9 %
W1 (unseen robots)
31
31
31
1.9 %
3.7 %
W3 (unseen combination)
108
108
108
2.6 %
4.6 %
W4 ( N=32 )
282
282
282
3.3 %
6.4 %
Appendix
Table 10: Violation onsets of OntoCBF with PD preceded by a conflict of an agent of the violated constraint, and base rates over agent-steps (window 1 s).
Configuration (two drones substituted)
States
Jointly valid
All per-agent QPs feasible
Valid where the design team is
Drone (design)
78837
94.43 %
92.44 %
74442/74442
LightDrone , inherited β=2.0
78837
94.57 %
92.56 %
74442/74442
HeavyDrone , β=2.0 (Naive)
78837
93.34 %
91.56 %
73589/74442
HeavyDrone , own β=1.0
64845
87.11 %
84.79 %
56485/61668
Appendix
Table 11: Joint validity (Definition 2 ) on the states visited by the design team in W0, with both drones substituted.
PD
MAPPO
No filter
Scenario
Viol.
Inf. %
Deliv.
Viol.
Inf. %
Deliv.
Viol. ep. %
W0 (design)
6.1
1.20
1.76
3.1
0.73
1.73
100
W1 (unseen robots)
5.6
1.15
1.64
3.3
0.75
1.69
100
W2 (unseen payload)
6.1
1.20
1.76
3.1
0.73
1.73
100
W3 (unseen combination)
16.0
1.62
1.43
9.0
0.95
1.43
100
W4 ( N=32 )
47.9
2.21
1.80
39.6
1.81
1.69
100
Appendix
Table 12: PD and MAPPO behind OntoCBF, 100 paired seeds per cell. MAPPO pools 3 runs trained in W0 only, so its cells hold 3 × 100 episodes, and the intervals in Section 6 resample training runs and then seeds. The columns give violating steps per episode (Viol.), infeasible QPs as a percentage of agent-steps with active constraints (Inf.) and deliveries per courier (Deliv.), and the last column gives the percentage of episodes with a violation for MAPPO without the filter.
Scenario
Run
Δ infeasible QPs / ep.
Δ violating steps / ep.
Δ deliveries / courier
W0 (design)
1
−23.9 [ −37.7 , −10.5 ]
−1.3 [ −5.3 , +2.7 ]
−0.01 [ −0.05 , +0.04 ]
2
−34.4 [ −47.6 , −21.6 ]
−4.4 [ −7.8 , −1.6 ]
+0.04 [ −0.01 , +0.08 ]
3
−32.0 [ −47.0 , −16.4 ]
−3.2 [ −7.1 , +0.7 ]
−0.12 [ −0.17 , −0.07 ]
W1 (unseen robots)
1
−26.0 [ −38.3 , −13.7 ]
−2.3 [ −5.7 , +1.0 ]
+0.01 [ −0.04 , +0.05 ]
2
−23.4 [ −36.0 , −10.9 ]
−2.8 [ −5.9 , +0.2 ]
+0.05 [ +0.01 , +0.10 ]
3
−27.3 [ −40.4 , −13.6 ]
−1.9 [ −5.5 , +1.5 ]
+0.08 [ +0.03 , +0.12 ]
Appendix
Table 13: MAPPO minus PD behind OntoCBF for each training run (W2 equals W0), with paired 95 % bootstrap intervals over the 100 seeds (10,000 resamples). Violating steps fall significantly in 4 of the 12 run–scenario cells.
Multi-robot Control Barrier Function (CBF) safety filters can become infeasible, but a failed quadratic program (QP) does not indicate why the conflict occurred or how to resolve it. To address this, we develop an exact feasibility certificate for multi-agent CBF filters with heterogeneous control-affine dynamics and convex input sets. The certificate quantifies a feasibility reserve by separating the demand imposed by safety constraints from the available actuator supply. This decomposition shows when CBF gain tuning or increased actuation can and cannot resolve infeasibility, and identifies the agents and interactions responsible for the conflict. We further propose an algorithm to optimally allocate shared safety constraints by maximizing the worst local feasibility margin, yielding a linear program for polyhedral input sets. In 320 paired closed-loop simulations, the proposed allocation reduces infeasible control steps from roughly 50% to 6.2%, and reduces safety-violating runs from 118/160 to 24/160. In addition, across 52 infeasibility events, the certificate identifies an interaction whose relaxation restores feasibility in 94% of cases.
Chandan Kumar Sah, Jishnu Keshavan
Department of Mechanical Engineering, Indian Institute of Science, Bengaluru 560012, India
Reinforcement learning (RL) can teach robot teams to navigate shared spaces, but trial and error causes collisions that real robots cannot afford. Control barrier function (CBF) safety filters prevent these collisions by correcting unsafe actions, yet in a team the filter becomes a centralized optimization whose number of constraints grows quadratically with the team size. We propose an approach that builds the filter's safety into the RL policy and examine how such a system behaves during filter-free deployment. However, the filter that keeps training safe also hides what the policy must learn: it masks the cost of unsafe actions, which actor-critic methods then overvalue, and it keeps near-collision states out of the training data. Our approach therefore learns only from executed safe actions with implicit Q-learning and distills the filter into the policy at synthetic near-collision states. We prove that filter distillation loss bounds the policy's constraint violation and hence how far agents can cross their safety margin, so errors never accumulate. When distillation states cover deployed states, the loss also bounds the collision probability at every step. Without a filter and acting only on local observations, our policies have no collisions in 1,800 test episodes with multiple agents. A policy trained with three agents transfers zero-shot to teams of twenty at constant per-agent computation and stays collision-free with up to eight agents. We test our policies on Crazyflie quadrotors with up to five flying at the same time.
Srikar Babu Gadipudi, Somayeh Sojoudi, Claire J. Tomlin
Safe decentralized navigation in heterogeneous robot teams is challenging because robots with different dynamics and control limits cannot satisfy identical collision-avoidance obligations equally well. Existing methods often rely on model-specific safety formulations or allocate avoidance responsibility without considering whether the assigned control is physically realizable, possibly leading to safety-constrained QP infeasibility and safety violations. We propose Capability-Aware Heterogeneous Control Barrier Functions (CA-HCBF), a decentralized framework for consistent safety enforcement and capability-aware responsibility allocation. We derive a unified operational-space second-order control-affine formulation for holonomic and nonholonomic mobile robots through canonical transformation and backstepping, addressing the relative-degree mismatch while preserving forward invariance of the pairwise safe set. Second, we quantify each robot's achievable control authority along the safety-critical direction using a support-function-based capability metric and use it to distribute pairwise avoidance responsibility. A feasibility-aware clipping mechanism further mitigates physically unrealizable responsibility assignments. Simulations with up to 30 heterogeneous robots and hardware experiments demonstrate high task completion with low safety violations and reduction in local QP infeasibility.
Joonkyung Kim, Yanze Zhang, Wenhao Luo +1
Department of Computer Science and Engineering, Texas A&M University · Department of Computer Science, University of Illinois Chicago