In cooperative multi-agent reinforcement studying (MARL), on account of its on-policy nature, coverage gradient (PG) strategies are usually believed to be much less pattern environment friendly than worth decomposition (VD) strategies, that are off-policy. Nevertheless, some current empirical research reveal that with correct enter illustration and hyper-parameter tuning, multi-agent PG can obtain surprisingly sturdy efficiency in comparison with off-policy VD strategies.
Why may PG strategies work so nicely? On this submit, we’ll current concrete evaluation to indicate that in sure eventualities, e.g., environments with a extremely multi-modal reward panorama, VD might be problematic and result in undesired outcomes. Against this, PG strategies with particular person insurance policies can converge to an optimum coverage in these circumstances. As well as, PG strategies with auto-regressive (AR) insurance policies can be taught multi-modal insurance policies.
Determine 1: totally different coverage illustration for the 4-player permutation sport.
CTDE in Cooperative MARL: VD and PG strategies
Centralized coaching and decentralized execution (CTDE) is a well-liked framework in cooperative MARL. It leverages world info for simpler coaching whereas holding the illustration of particular person insurance policies for testing. CTDE might be applied through worth decomposition (VD) or coverage gradient (PG), main to 2 several types of algorithms.
VD strategies be taught native Q networks and a mixing perform that mixes the native Q networks to a world Q perform. The blending perform is often enforced to fulfill the Particular person-International-Max (IGM) precept, which ensures the optimum joint motion might be computed by greedily selecting the optimum motion domestically for every agent.
Against this, PG strategies immediately apply coverage gradient to be taught a person coverage and a centralized worth perform for every agent. The worth perform takes as its enter the worldwide state (e.g., MAPPO) or the concatenation of all of the native observations (e.g., MADDPG), for an correct world worth estimate.
The permutation sport: a easy counterexample the place VD fails
We begin our evaluation by contemplating a stateless cooperative sport, particularly the permutation sport. In an $N$-player permutation sport, every agent can output $N$ actions ${ 1,ldots, N }$. Brokers obtain $+1$ reward if their actions are mutually totally different, i.e., the joint motion is a permutation over $1, ldots, N$; in any other case, they obtain $0$ reward. Observe that there are $N!$ symmetric optimum methods on this sport.
Determine 2: the 4-player permutation sport.
Allow us to concentrate on the 2-player permutation sport for our dialogue. On this setting, if we apply VD to the sport, the worldwide Q-value will factorize to
[Q_textrm{tot}(a^1,a^2)=f_textrm{mix}(Q_1(a^1),Q_2(a^2)),]
the place $Q_1$ and $Q_2$ are native Q-functions, $Q_textrm{tot}$ is the worldwide Q-function, and $f_textrm{combine}$ is the blending perform that, as required by VD strategies, satisfies the IGM precept.
Determine 3: high-level instinct on why VD fails within the 2-player permutation sport.
We formally show that VD can’t characterize the payoff of the 2-player permutation sport by contradiction. If VD strategies have been capable of characterize the payoff, we might have
[Q_textrm{tot}(1, 2)=Q_textrm{tot}(2,1)=1 qquad textrm{and} qquad Q_textrm{tot}(1, 1)=Q_textrm{tot}(2,2)=0.]
Nevertheless, if both of those two brokers have totally different native Q values, e.g. $Q_1(1)> Q_1(2)$, then in response to the IGM precept, we should have
[1=Q_textrm{tot}(1,2)=argmax_{a^2}Q_textrm{tot}(1,a^2)>argmax_{a^2}Q_textrm{tot}(2,a^2)=Q_textrm{tot}(2,1)=1.]
In any other case, if $Q_1(1)=Q_1(2)$ and $Q_2(1)=Q_2(2)$, then
[Q_textrm{tot}(1, 1)=Q_textrm{tot}(2,2)=Q_textrm{tot}(1, 2)=Q_textrm{tot}(2,1).]
Consequently, worth decomposition can’t characterize the payoff matrix of the 2-player permutation sport.
What about PG strategies? Particular person insurance policies can certainly characterize an optimum coverage for the permutation sport. Furthermore, stochastic gradient descent can assure PG to converge to certainly one of these optima underneath delicate assumptions. This implies that, though PG strategies are much less fashionable in MARL in contrast with VD strategies, they are often preferable in sure circumstances which can be widespread in real-world functions, e.g., video games with a number of technique modalities.
We additionally comment that within the permutation sport, with a view to characterize an optimum joint coverage, every agent should select distinct actions. Consequently, a profitable implementation of PG should be sure that the insurance policies are agent-specific. This may be completed by utilizing both particular person insurance policies with unshared parameters (known as PG-Ind in our paper), or an agent-ID conditioned coverage (PG-ID).
PG outperform greatest VD strategies on fashionable MARL testbeds
Going past the straightforward illustrative instance of the permutation sport, we lengthen our examine to fashionable and extra reasonable MARL benchmarks. Along with StarCraft Multi-Agent Problem (SMAC), the place the effectiveness of PG and agent-conditioned coverage enter has been verified, we present new leads to Google Analysis Soccer (GRF) and multi-player Hanabi Problem.
Determine 4: (left) profitable charges of PG strategies on GRF; (proper) greatest and common analysis scores on Hanabi-Full.
In GRF, PG strategies outperform the state-of-the-art VD baseline (CDS) in 5 eventualities. Apparently, we additionally discover that particular person insurance policies (PG-Ind) with out parameter sharing obtain comparable, typically even increased profitable charges, in comparison with agent-specific insurance policies (PG-ID) in all 5 eventualities. We consider PG-ID within the full-scale Hanabi sport with various numbers of gamers (2-5 gamers) and evaluate them to SAD, a powerful off-policy Q-learning variant in Hanabi, and Worth Decomposition Networks (VDN). As demonstrated within the above desk, PG-ID is ready to produce outcomes corresponding to or higher than the perfect and common rewards achieved by SAD and VDN with various numbers of gamers utilizing the identical variety of atmosphere steps.
Past increased rewards: studying multi-modal habits through auto-regressive coverage modeling
Apart from studying increased rewards, we additionally examine how you can be taught multi-modal insurance policies in cooperative MARL. Let’s return to the permutation sport. Though we’ve proved that PG can successfully be taught an optimum coverage, the technique mode that it lastly reaches can extremely depend upon the coverage initialization. Thus, a pure query shall be:
Can we be taught a single coverage that may cowl all of the optimum modes?
Within the decentralized PG formulation, the factorized illustration of a joint coverage can solely characterize one explicit mode. Subsequently, we suggest an enhanced strategy to parameterize the insurance policies for stronger expressiveness — the auto-regressive (AR) insurance policies.
Determine 5: comparability between particular person insurance policies (PG) and auto-regressive insurance policies (AR) within the 4-player permutation sport.
Formally, we factorize the joint coverage of $n$ brokers into the type of
[pi(mathbf{a} mid mathbf{o}) approx prod_{i=1}^n pi_{theta^{i}} left( a^{i}mid o^{i},a^{1},ldots,a^{i-1} right),]
the place the motion produced by agent $i$ relies upon by itself remark $o_i$ and all of the actions from earlier brokers $1,dots,i-1$. The auto-regressive factorization can characterize any joint coverage in a centralized MDP. The solely modification to every agent’s coverage is the enter dimension, which is barely enlarged by together with earlier actions; and the output dimension of every agent’s coverage stays unchanged.
With such a minimal parameterization overhead, AR coverage considerably improves the illustration energy of PG strategies. We comment that PG with AR coverage (PG-AR) can concurrently characterize all optimum coverage modes within the permutation sport.
Determine: the heatmaps of actions for insurance policies discovered by PG-Ind (left) and PG-AR (center), and the heatmap for rewards (proper); whereas PG-Ind solely converge to a particular mode within the 4-player permutation sport, PG-AR efficiently discovers all of the optimum modes.
In additional advanced environments, together with SMAC and GRF, PG-AR can be taught fascinating emergent behaviors that require sturdy intra-agent coordination which will by no means be discovered by PG-Ind.
Determine 6: (left) emergent habits induced by PG-AR in SMAC and GRF. On the 2m_vs_1z map of SMAC, the marines maintain standing and assault alternately whereas making certain there is just one attacking marine at every timestep; (proper) within the academy_3_vs_1_with_keeper state of affairs of GRF, brokers be taught a “Tiki-Taka” model habits: every participant retains passing the ball to their teammates.
Discussions and Takeaways
On this submit, we offer a concrete evaluation of VD and PG strategies in cooperative MARL. First, we reveal the limitation on the expressiveness of fashionable VD strategies, exhibiting that they may not characterize optimum insurance policies even in a easy permutation sport. Against this, we present that PG strategies are provably extra expressive. We empirically confirm the expressiveness benefit of PG on fashionable MARL testbeds, together with SMAC, GRF, and Hanabi Problem. We hope the insights from this work may gain advantage the neighborhood in the direction of extra basic and extra highly effective cooperative MARL algorithms sooner or later.
This submit relies on our paper in joint with Zelai Xu: Revisiting Some Widespread Practices in Cooperative Multi-Agent Reinforcement Studying (paper, web site).
